Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
THIS MONTH’S DIGITAL DELIVERY OF IEEE COMMUNICATIONS MAGAZINE SUPPORTED BY:
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Contents | Zoom in | Zoom out
For navigation instructions please click here
Search Issue | Next Page
IEEE
July 2011, Vol. 49, No. 7
www.comsoc.org
MAGAZINE C l ria to Tu c 1 So 6 om IPv age P
IEEE Standards for Wireless
e Se
Network and Service Management
ee Fr
Future Internet Architectures
Free ComSoc Articles on 3GPP See Page 3
A Publication of the IEEE Communications Society
Contents | Zoom in | Zoom out
For navigation instructions please click here
Search Issue | Next Page
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Introducing the Cisco® Carrier Packet Transport (CPT) System, the industry’s first standards-based, Packet Optical Transport System (P-OTS) that unifies packet and transport technologies using Multiprotocol Label Switching-Transport Profile (MPLS-TP). Enhance your service offerings and increase profitability Learn more about the Cisco Carrier Packet Transport System and view the Current Analysis report at www.cisco.com/go/cpt.
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
TM
A
BEMaGS F
____________________________________________________
New 2011 tutorial
BUILDING A COMPREHENSIVE
IPv6
TRANSITION
STRATEGY
To maintain business continuity, service providers must now accelerate their transition to Internet Protocol Version 6 (IPv6). Deploying IPv6 requires a well-constructed network design, a detailed deployment plan, and thorough testing to ensure compatibility with existing network characteristics. Depending on the primary drivers for IPv6 deployment and the state of the current IPv4 network, Service Providers may choose different approaches for integrating IPv6 into their networks. This tutorial is intended help Service Providers build a comprehensive IPv6 transition strategy.
C ris Metz, etz, Cisco Chris
FREE ACCESS SPONSORED BY
Brought to you by
For other sponsor opportunities, please contact Eric Levine, Associate Publisher Phone: 212-705-8920, E-mail: ______________
[email protected]
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Director of Magazines Andrzej Jajszczyk, AGH U. of Sci. & Tech. (Poland)
A
BEMaGS F
IEEE
Editor-in-Chief Steve Gorshe, PMC-Sierra, Inc. (USA) Associate Editor-in-Chief Sean Moore, Centripetal Networks (USA) Senior Technical Editors Tom Chen, Swansea University (UK) Nim Cheung, ASTRI (China) Nelson Fonseca, State Univ. of Campinas (Brazil) Peter T. S. Yum, The Chinese U. Hong Kong (China) Technical Editors Sonia Aissa, Univ. of Quebec (Canada) Mohammed Atiquzzaman, U. of Oklahoma (USA) Paolo Bellavista, DEIS (Italy) Tee-Hiang Cheng, Nanyang Tech. U. (Rep. Singapore) Sudhir S. Dixit, Hewlett-Packard Labs India (India) Stefano Galli, ASSIA, Inc. (USA) Joan Garcia-Haro, Poly. U. of Cartagena (Spain) Admela Jukan, Tech. Univ. Carolo-Wilhelmina zu Braunschweig (Germany) Vimal Kumar Khanna, mCalibre Technologies (India) Janusz Konrad, Boston University (USA) Deep Medhi, Univ. of Missouri-Kansas City (USA) Nader F. Mir, San Jose State Univ. (USA) Amitabh Mishra, Johns Hopkins University (USA) Seshradi Mohan, University of Arkansas (USA) Glenn Parsons, Ericsson Canada (Canada) Joel Rodrigues, Univ. of Beira Interior (Portugal) Jungwoo Ryoo, The Penn. State Univ.-Altoona (USA) Hady Salloum, Stevens Institute of Tech. (USA) Antonio Sánchez Esguevillas, Telefonica (Spain) Dan Keun Sung, Korea Adv. Inst. Sci. & Tech. (Korea) Danny Tsang, Hong Kong U. of Sci. & Tech. (Japan) Chonggang Wang, InterDigital Commun., LLC (USA) Alexander M. Wyglinski, Worcester Poly. Institute (USA) Series Editors Ad Hoc and Sensor Networks Edoardo Biagioni, U. of Hawaii, Manoa (USA) Silvia Giordano, Univ. of App. Sci. (Switzerland) Automotive Networking and Applications Wai Chen, Telcordia Technologies, Inc (USA) Luca Delgrossi, Mercedes-Benz R&D N.A. (USA) Timo Kosch, BMW Group (Germany) Tadao Saito, University of Tokyo (Japan) Consumer Communicatons and Networking Madjid Merabti, Liverpool John Moores U. (UK) Mario Kolberg, University of Sterling (UK) Stan Moyer, Telcordia (USA) Design & Implementation Sean Moore, Avaya (USA) Salvatore Loreto, Ericsson Research (Finland) Integrated Circuits for Communications Charles Chien (USA) Zhiwei Xu, SST Communication Inc. (USA) Stephen Molloy, Qualcomm (USA) Network and Service Management Series George Pavlou, U. of Surrey (UK) Aiko Pras, U. of Twente (The Netherlands) Networking Testing Series Yingdar Lin, National Chiao Tung University (Taiwan) Erica Johnson, University of New Hampshire (USA) Tom McBeath, Spirent Communications Inc. (USA) Eduardo Joo, Empirix Inc. (USA) Topics in Optical Communications Hideo Kuwahara, Fujitsu Laboratories, Ltd. (Japan) Osman Gebizlioglu, Telcordia Technologies (USA) John Spencer, Optelian (USA) Vijay Jain, Verizon (USA) Topics in Radio Communications Joseph B. Evans, U. of Kansas (USA) Zoran Zvonar, MediaTek (USA) Standards Yoichi Maeda, NTT Adv. Tech. Corp. (Japan) Mostafa Hashem Sherif, AT&T (USA) Columns Book Reviews Piotr Cholda, AGH U. of Sci. & Tech. (Poland) History of Communications Steve Weinsten (USA) Regulatory and Policy Issues J. Scott Marcus, WIK (Germany) Jon M. Peha, Carnegie Mellon U. (USA) Technology Leaders' Forum Steve Weinstein (USA) Very Large Projects Ken Young, Telcordia Technologies (USA) Publications Staff Joseph Milizzo, Assistant Publisher Eric Levine, Associate Publisher Susan Lange, Online Production Manager Jennifer Porcello, Production Specialist Catherine Kemelmacher, Associate Editor
®
2
Communications IEEE
MAGAZINE July 2011, Vol. 49, No. 7
www.comsoc.org/~ci FUTURE INTERNET ARCHITECTURES: DESIGN AND DEPLOYMENT PERSPECTIVES GUEST EDITORS: RAJ JAIN, ARJAN DURRESI, AND SUBHARTHI PAUL
24 26
GUEST EDITORIAL
38
LOCI OF COMPETITION FOR FUTURE INTERNET ARCHITECTURES
44
BIOLOGICAL PRINCIPLES FOR FUTURE INTERNET ARCHITECTURE DESIGN
54
ENABLING FUTURE INTERNET RESEARCH: THE FEDERICA CASE
62
A SURVEY OF THE RESEARCH ON FUTURE INTERNET ARCHITECTURES The current Internet, which was designed over 40 years ago, is facing unprecedented challenges in many aspects, especially in the commercial context. JIANLI PAN, SUBHARTHI PAUL, AND RAJ JAIN Designing for competition is an important consideration for the design of future Internet architectures. Network architects should systematically consider the loci of competition in any proposed network architecture. JOHN CHUANG The Internet has evolved to accommodate unexpected diversity in services and applications. This trend will continue. The architecture of the new-generation Internet must be designed in a dynamic, modular, and adaptive way. Features like these can often be observed in biological processes that serve as inspiration for designing new cooperative architectural concepts. SASITHARAN BALASUBRAMANIAM, KENJI LEIBNITZ, PIETRO LIO’, DMITRI BOTVICH, AND MASAYUKI MURATA The authors provide a comprehensive overview of the state-of-the-art research projects that have been using the virtual infrastructure slices of FEDERICA in order to validate their research concepts, even when they are disruptive to the testbed’s infrastructure, to obtain results in realistic network environments. PETER SZEGEDI, JORDI FERRER RIERA, JOAN A. GARCIA-ESPIN, MARKUS HIDELL, PETER SJÖDIN, PEHR SÖDERMAN, MARCO RUFFINI, DONAL O’MAHONY, ANDREA BIANCO, LUCA GIRAUDO, MIGUEL PONCE DE LEON, GEMMA POWER, CRISTINA CERVELLO-PASTOR, VICTOR LOPEZ, AND SUSANNE NAEGELE-JACKSON
CONTENT, CONNECTIVITY, AND CLOUD: INGREDIENTS FOR THE NETWORK OF THE FUTURE A new network architecture for the Internet needs ingredients from three approaches: information-centric networking, cloud computing integrated with networking, and open connectivity. BENGT AHLGREN, PEDRO A. ARANDA, PROSPER CHEMOUIL, SARA OUESLATI, LUIS M. CORREIA, HOLGER KARL, MICHAEL SÖLLNER, AND ANNIKKI WELIN
71
PEARL: A PROGRAMMABLE VIRTUAL ROUTER PLATFORM The authors present the design and implementation of PEARL, a programmable virtual router platform with relatively high performance. GAOGANG XIE, PENG HE, HONGTAO GUAN, ZHENYU LI, YINGKE XIE, LAYONG LUO, JIANHUA ZHANG, YONGGONG WANG, AND KAVÉ SALAMATIAN
TOPICS IN NETWORK AND SERVICE MANAGEMENT SERIES EDITORS: GEORGE PAVLOU AND AIKO PRAS
78 80
SERIES EDITORIAL
88
NETWORK RESILIENCE: A SYSTEMATIC APPROACH
TOWARD DECENTRALIZED PROBABILISTIC MANAGEMENT The authors discuss the potential of decentralized probabilistic management and its impact on management operations, and illustrate the paradigm by three example solutions for real-time monitoring and anomaly detection. ALBERTO GONZALEZ PRIETO, DANIEL GILLBLAD, REBECCA STEINERT, AVI MIRON Aspects of network resilience, such as the application of fault-tolerant systems techniques to optical switching, have been studied and applied to great effect. However, networks — and the Internet in particular — are still vulnerable PAUL SMITH, DAVID HUTCHISON, JAMES P. G. STERBENZ, MARCUS SCHÖLLER, ALI FESSI, MERKOURIS KARALIOPOULOS, CHIDUNG LAC, AND BERNHARD PLATTNER
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
http://www.comsoc.org/te g/techfocus
presented by IEEE Communications Society
BEMaGS F
A FREE ONLINE OFFER from the IEEE COMMUNICATIONS DIGITAL LIBRARY
3GPP
TECHNOLOGY FOCUS
A
LIMITED TIME OFFER!
Enjoy free access to full content from IEEE Communications Society publications and conferences. This is the only site with free access from the IEEE Communications Digital Library on 3GPP. Conference papers and publication articles were originally presented at IEEE Communications Society conferences including IEEE GLOBECOM, ICC, PIMRC, DYSPAN, WCNC, and IEEE Communications Society Magazines and Journals. Free 3GPP articles and conference papers include: Synchronization and cell search in 3GPP LTE systems QoS Architecture for the 3GPP IETF-Based Evolved Packet Core Network-based mobility management in the evolved 3GPP core network 3GPP Machine-to-Machine Communications Handover between mobile WiMAX and 3GPP UTRAN Assessing 3GPP LTE-Advanced as IMT-Advanced Technology Robust channel estimation and detection in 3GPP-LTE 3GPP LTE downlink system performance Downlink MIMO with frequency-domain packet scheduling for 3GPP LTE Coexistence studies for 3GPP LTE with other mobile systems QoS control in the 3GPP evolved packet system Combating timing asynchronism in relay transmission for 3GPP LTE uplink …plus many more papers from the IEEE Communications Society; all with limited time free access! Go to http://www.comsoc.org/techfocus
free access compliments of
For other sponsor opportunities, please contact Eric Levine, Associate Publisher Phone: 212-705-8920, E-mail:
[email protected] ______________
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
2011 Communications Society Elected Officers Byeong Gi Lee, President Vijay Bhargava, President-Elect Mark Karol, VP–Technical Activities Khaled B. Letaief, VP–Conferences Sergio Benedetto, VP–Member Relations Leonard Cimini, VP–Publications Members-at-Large Class of 2011 Robert Fish, Joseph Evans Nelson Fonseca, Michele Zorzi Class of 2012 Stefano Bregni, V. Chan Iwao Sasase, Sarah K. Wilson Class of 2013 Gerhard Fettweis, Stefano Galli Robert Shapiro, Moe Win 2011 IEEE Officers Moshe Kam, President Gordon W. Day, President-Elect Roger D. Pollard, Secretary Harold L. Flescher, Treasurer Pedro A. Ray, Past-President E. James Prendergast, Executive Director Nim Cheung, Director, Division III IEEE COMMUNICATIONS MAGAZINE (ISSN 01636804) is published monthly by The Institute of Electrical and Electronics Engineers, Inc. Headquarters address: IEEE, 3 Park Avenue, 17th Floor, New York, NY 10016-5997, USA; tel: +1-212705-8900; http://www.comsoc.org/ci. Responsibility for the contents rests upon authors of signed articles and not the IEEE or its members. Unless otherwise specified, the IEEE neither endorses nor sanctions any positions or actions espoused in IEEE Communications Magazine. ANNUAL SUBSCRIPTION: $27 per year print subscription. $16 per year digital subscription. Non-member print subscription: $400. Single copy price is $25.
98
104
AND
REPRINT
POSTMASTER:
Send address changes to IEEE Communications Magazine, IEEE, 445 Hoes Lane, Piscataway, NJ 08855-1331. GST Registration No. 125634188. Printed in USA. Periodicals postage paid at New York, NY and at additional mailing offices. Canadian Post International Publications Mail (Canadian Distribution) Sales Agreement No. 40030962. Return undeliverable Canadian addresses to: Frontier, PO Box 1051, 1031 Helena Street, Fort Eire, ON L2A 6C7
SUBMISSIONS: The magazine welcomes tutorial or survey articles that span the breadth of communications. Submissions will normally be approximately 4500 words, with few mathematical formulas, accompanied by up to six figures and/or tables, with up to 10 carefully selected references. Electronic submissions are preferred, and should be sumitted through Manuscript Central http://mc.manuscriptcentral.com/commag-ieee. Instructions can be found at the following: http://dl.com______ soc.org/livepubs/ci1/info/sub_guidelines.html. For further information contact Sean Moore, Associate Editor-inChief (
[email protected]). __________ All submissions will be peer reviewed.
VLANs are widely used in today’s enterprise networks to improve Ethernet scalability and support network policies. However, manuals and textbooks offer very little information about how VLANs are actually used in practice. MINLAN YU, JENNIFER REXFORD, XIN SUN, SANJAY RAO, AND NICK FEAMSTER
TOWARD FINE-GRAINED TRAFFIC CLASSIFICATION The authors propose a fine-grained traffic classification scheme based on the analysis of existing classification methodologies. BYUNGCHUL PARK, JAMES WON-KI HONG, AND YOUNG J. WON
SERIES EDITORIAL IEEE 802.15.3C: THE FIRST IEEE WIRELESS STANDARD FOR DATA RATES OVER 1 GB/S The authors explain the important features of IEEE 802.15.3c, the first wireless standard from IEEE in the 60-GHz (millimeter wave) band and its development. TUNCER BAYKAS, CHIN-SEAN SUM, ZHOU LAN, JUNYI WANG, M. AZIZUR RAHMAN, HIROSHI HARADA, AND SHUZO KATO
OVERVIEW OF FEMTOCELL SUPPORT IN ADVANCED WIMAX SYSTEMS
132
SMART UTILITY NETWORKS IN TV WHITE SPACE
140
ADVANCES IN MODE-STIRRED REVERBERATION CHAMBERS FOR WIRELESS COMMUNICATION PERFORMANCE EVALUATION
The authors provide an update on novel concepts and mechanisms for femtocell support in the network architecture and air interface that have been adopted into the WiMAX Forum network specifications and the IEEE 802.16m specification. YING LI, ANDREAS MAEDER, LINGHANG FAN, ANSHUMAN NIGAM, AND JOEY CHOU The authors present an overview of the background, technology, regulation, and standardization in the course of deploying smart utility networks (SUNs) in TV white space (TVWS) communications, two wireless technologies currently receiving overwhelming interest in the wireless industry and academia. CHIN-SEAN SUM, HIROSHI HARADA, FUMIHIDE KOJIMA, ZHOU LAN, AND RYUHEI FUNADA
ACCEPTED FROM OPEN CALL The authors highlight recent advances in the development of second-generation mode-stirred chambers for wireless communications performance evaluation. MIGUEL Á. GARCIA-FERNANDEZ, JUAN D. SANCHEZ-HEREDIA, ANTONIO M. MARTINEZ-GONZALEZ, DAVID A. SANCHEZ-HERNANDEZ, AND JUAN F. VALENZUELA-VALDÉS
148
SYSTEM-LEVEL SIMULATION METHODOLOGY AND PLATFORM FOR MOBILE CELLULAR SYSTEMS The authors propose a general unified simulation methodology for different cellular systems. LI CHEN, WENWEN CHEN, BIN WANG, XIN ZHANG, HONGYANG CHEN, AND DACHENG YANG
156
LAYER 3 WIRELESS MESH NETWORKS: MOBILITY MANAGEMENT ISSUES
164
PREAMBLE DESIGN, SYSTEM ACQUISITION, AND DETERMINATION IN MODERN OFDMA CELLULAR COMMUNICATIONS: AN OVERVIEW
SUBSCRIPTIONS, orders, address changes — IEEE Service Center, 445 Hoes Lane, Piscataway, NJ 08855-1331, USA; tel: +1-732-981-0060; e-mail:
[email protected]. ____________ ADVERTISING: Advertising is accepted at the discretion of the publisher. Address correspondence to: Advertising Manager, IEEE Communications Magazine, 3 Park Avenue, 17th Floor, New York, NY 10016.
A SURVEY OF VIRTUAL LAN USAGE IN CAMPUS NETWORKS
122
PERMISSIONS:
Abstracting is permitted with credit to the source. Libraries are permitted to photocopy beyond the limits of U.S. Copyright law for private use of patrons: those post-1977 articles that carry a code on the bottom of the first page provided the per copy fee indicated in the code is paid through the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. For other copying, reprint, or republication permission, write to Director, Publishing Services, at IEEE Headquarters. All rights reserved. Copyright © 2011 by The Institute of Electrical and Electronics Engineers, Inc.
Wireless Mesh Networks (WMNs) may be broadly classified into two categories: Layer 2 and Layer 3 WMNs. The authors focus on the Layer 3 WMN, which provides the same service interfaces and functionalities to the conventional mobile host (MH) as the conventional wireless local area network. KENICHI MASE
The wide choices of deployment parameters in next generation wireless communication systems present significant challenges in preamble and system acquisition design. The authors address these challenges, as well as the solutions provided by the next generation wireless standards, MICHAEL MAO WANG, AVNEESH AGRAWAL, AAMOD KHANDEKAR, AND SANDEEP AEDUDODLA
176
EVALUATING STRATEGIES FOR EVOLUTION OF PASSIVE OPTICAL NETWORKS
185
ON ASSURING END-TO-END AND A POSSIBLE SOLUTION
The authors study the requirements for optimal migration toward higher bandwidth per user, and examine scenarios and cost-effective solutions for PON evolution. MARILET DE ANDRADE, GLEN KRAMER, LENA WOSINSKA, JIAJIA CHEN, SEBASTIÀ SALLENT, AND BISWANATH MUKHERJEE
IEEE
QOE IN NEXT GENERATION NETWORKS: CHALLENGES
The authors discuss challenges and a possible solution for optimizing end-to-end QoE in Next Generation Networks. JINGJING ZHANG AND NIRWAN ANSARI President’s Page Conference Report/ICC 2011 Conference Calendar Society News
Communications
F
STANDARDS SERIES EDITORS: MOSTAFA HASHEM SHERIF AND YOICHI MAEDA
112 114
in-Chief, Steve Gorshe, PMC-Sierra, Inc., 10565 S.W. Nimbus Avenue, Portland, OR 97223; tel: +(503)
[email protected]. 7440, e-mail: ______________
4
BEMaGS
RECENT IEEE STANDARDS FOR WIRELESS DATA COMMUNICATIONS
EDITORIAL CORRESPONDENCE: Address to: Editor-
COPYRIGHT
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
6 12 16 17
New Products Global Communications Newsletter Advertisers’ Index
18 19 192
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
TECHNICAL CONFERENCE March 4–8, 2012 EXPOSITION March 6–8, 2012 LOS ANGELES CONVENTION CENTER Los Angeles, California, USA
CALL FOR PAPERS SUBMISSION DEADLINE: OCTOBER 6, 2011, 12:00 P.M. NOON EDT (16:00 GMT)
SUBMIT YOUR RESEARCH TO THE WORLD’S LARGEST CONFERENCE FOR ADVANCING OPTICAL SOLUTIONS IN TELECOM, DATACOM, COMPUTING AND MORE
Scan QR Code with your smartphone for more information
STUDENTS: Enter the Corning Outstanding Student Paper Competition for Your Chance to Win $1,500!
WWW.OFCNFOEC.ORG/SUBMISSIONS ____________________________
Communications IEEE
SPONSORED BY
NON-FINANCIAL TECHNICAL CO-SPONSOR
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
THE PRESIDENT’S PAGE
“GLOBAL COMSOC” EMBRACING THE GLOBE
T
toward globally balanced representation by he “Golden Triangle” vision, as introMembers-at-Large (MaL) in the BoG. duced in the January 2010 President’s In this President’s Page we will introduce Page, supports transformation of our IEEE ComSoc’s efforts to achieve our globalization Communications Society (ComSoc) into a goal, the programs we support for our global truly global, vital and high-value professional membership, and how we plan to enhance society. The three vertices of the “Golden Triour existing successful programs to become angle,” namely Globalization, Young Leaders, more global. The contributors to this page and Industry, are the three fundamental coninclude: Shri Goyal, ComSoc Director-Memcepts of transformation. Globalization enables bership Programs Development; Roberto utilization of the best talent, education, trainSaracco, Director-Societal Relations; Naoaki ing and cultural values among our members Yamanaka, Director-Asia Pacific Region; from around the world. Young Leaders are Tariq Durrani, Director-Europe, Middle the future of our Society. Interestingly, the East, and Africa Region; Jose David Cely, rapidly growing countries, fueling the drive BYEONG GI LEE Director-Latin Amerifor Globalization, have ca Region; and Gabriel higher percentages of Jakobson, Directortheir work force underNorth America Region. 30 years of age. IndusShri Goyal received try implements techhis M.Sc. degree in nology in products and Electronics from Allaservices, making these habad University in available to user comIndia and the Ph.D. in munities around the Electrical Engineering world, including rurual from North Carolina areas and the expandState University. Shri ing consumer base worked 25 plus years around the world. All at GTE (now Verithree vertices in the zon) Laboratories and “Golden Triangle,” was Dean of the Coltherefore, center around SHRI GOYAL ROBERTO SARACCO NAOAKI YAMANAKA lege of Technology & Globalization in the Management in St development and operPetersburg, Florida ations of our Communiuntil recently. Shri has cations Society. organized numerous Recognizing its importglobal symposia, inance, special emphasis cluding NOMS and has long been placed on IM. Shri served as the globalization in the Director of Meetings Communication Sociand Conferences (2004ety. ComSoc has been 6). Currently he is establishing its global ComSoc’s Director of footprint since at least Membership Programs the 1990s. In 1994, Development, embracMaurizio Decina, the ing activities that serve first non-U.S. based our members and ComSoc President, TARIQ DURRANI JOSE DAVID CELY GABRIEL JAKOBSON Chapters worldwide. coined the phrase Shri is an IEEE Fellow. “Global CommunicaRoberto Saracco is the Director of the Telecom Italia tions Society” to concisely embody the future direction of Future Centre, where he is leading a group of researchers in ComSoc. Since then ComSoc has been transforming from a the analysis of the impact of technology evolution on the primarily US-centric Society (before the Bell System divestitelecommunications business. He has participated in a number ture in 1984) into a global Society of today with a majority of of EU groups, including the 2020 Visionary Group and the members residing outside the US. Internet 2020 Group. He is a longstanding member of ComSoc’s globalization has been successful in membership, IEEE/COMSOC and has volunteered in several positions. publications, conferences, and technical activities. In parallel Currently he is the Director of Sister and Related Societies with that, we have also been promoting globalization in sharand the Chair ComSoc’s 2020 Committee. ing ComSoc’s leadership roles among different regional memNaoaki Yamanaka received his B.E., M.E., and Ph.D. bers. ComSoc’s Board of Governors (BoG) includes four degrees from Keio University, Japan. He joined Nippon TeleRegional Directors, and Presidents and Vice Presidents were graph and Telephone Corporation (NTT) in 1983, where he elected from all four ComSoc Regions. We are also working
6
Communications IEEE
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
THE PRESIDENT’S PAGE COMSOC’S GLOBALIZATION INITIATIVES
F IGURE 1. Geographic trends in ComSoc membership 20002010. conducted research and development on high-speed switching systems and technologies. Currently he is a Professor in the Information and Computer Science Department of Keio University. Yamanaka is ComSoc’s Director of the Asia Pacific Region and a Board member of the IEEE CPMT Society. He is a Fellow of the IEEE and a Fellow of the IEICE. He has published more than 122 journal and transaction articles, 82 international conference papers, and 174 patents, including 17 international patents. Tariq Durrani has been a Research Professor in Electronic & Electrical Engineering, University of Strathclyde, Glasgow, UK, since 1982. He was Department Head in 1990-94, and University Deputy Principal (Provost equivalent) in 2000-06. Tariq has held various visiting appointments around the world, including at Princeton University and the University of Southern California. He has authored/co-authored more than 350 papers and six books. His research interests cover Wireless Communications Systems and Signal Processing. He served as President of the IEEE Signal Processing Society (1994-95) and President of the IEEE Engineering Management Society (2006-07). Currently he is ComSoc’s Regional Director for Europe, Middle East and Africa, and the IEEE Vice President for Educational Activities (also chairing the IEEE Educational Activities Board). Jose David Cely C. received his B.S. degree from the Universidad Distrital Francisco Jose de Caldas in Bogota, Colombia. He worked for the Telecommunications Research Center of Colombia CINTEL, TELECOM Colombia, and the Universidad Catolica de Colombia. Currently he works for Universidad Distrital. He has served in several appointed and elected positions in IEEE and ComSoc, and actively participated in organizing LATINCOM 2009 and LATINCOM 2010. Currently he is ComSoc’s Director of the Latin America Region. He received ComSoc’s Latin America Region Distinguished Service Award in 2009 and the IEEE MGA’s Achievement Award in 2010. Gabriel Jakobson is Chief Scientist of Altusys Corp., a consulting company specializing in situation management technologies for defense, cyber security, and enterprise management applications. He received his Ph.D. in Computer Science from the Institute of Cybernetics, Estonia, and an Honorary Degree of Doctor Honorius Causa from Tallinn University of Technology, Estonia. Within ComSoc Gabe is a Distinguished Lecturer who has given lectures in more than 20 countries, the Director of the North America Region, the Vice-Chair of the Tactical Communications and Operations Technical Committee, and the Chair of the Sub-Committee on Situation Management.
Rapid globalization of communications and information technology services has been changing the demographics of ComSoc’s membership. Over the past 10 years the proportion of ComSoc membership from outside of its North America Region has increased from 44% to more than 59% (see Figure 1). It is natural that fast developing regions are assuming an increasingly heavier role in various ComSoc activities. As a world-wide organization ComSoc operates four global regions: North America Region (NAR), Latin America Region (LAR), Europe/Middle East/Africa Region (EMEAR), and Asia Pacific Region (APR), with a Regional Director (RD) representing each region. Our over 50,000 members in the four ComSoc Regions find their local technical home in 207 ComSoc Chapters. Our global members are served by our ComSoc headquarters in New York and a remote office in Singapore, collocated with the IEEE Singapore office. More satellite offices are being pursued in China, Russia, and India. Over the years, ComSoc has made efforts to engage members from all regions in its leadership and governance. Today various operational units in ComSoc make “open calls” for appointments when leadership positions are available and encourage all Regions to nominate eligible candidates. ComSoc has had a fairly global representation in its leadership positions (e.g. in boards, committees, and councils) and at the “grass roots” throughout its conferences, publications, and technical activities. In addition, ComSoc has been eagerly seeking, through its ad hoc Nomination and Election Process Committee, improved methods to encourage election of Members-at-Large to the Board of Governors (BoG) better in line with membership demographics.
GOLD Meeting at Globecom 2010 in Miami, Florida. The young professionals are the future of ComSoc. The Graduates of the Last Decade (GOLD) program, initiated by the IEEE Member and Geographic Activities Board (MGAB), provides an excellent framework for engaging young professional communities. ComSoc works closely with the IEEE MGAB and supports the GOLD program through members it appointes to its key councils and boards. We also hold GOLD sessions at our flagship and portfolio conferences, making special efforts to encourage young members’ participation. Through our mentoring, young members can take more significant leadership roles. Recently at IEEE GLOBECOM 2010 in Miami, we organized a special event by engaging young members from all four ComSoc Regions (see photo above). In this special session, they shared their experiences and discussed pros and cons about jobs and careers in academia and industry.
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
7
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
THE PRESIDENT’S PAGE GLOBALIZED MEMBER SERVICES AND PROGRAMS ComSoc offers a wide variety of programs to serve its members. Increasingly, these programs are being customized and tailored to serve the specific needs of local Regions and countries. Some key member services and programs are highlighted in the following sections. Chapter Funding and Recognition: ComSoc chapters have been growing steadily in numbers and activities. Ten new chapters were formed in 2010, including several in the developing countries in the Asia Pacific Region. Each chapter is funded on a yearly basis with the amount depending on the previous year’s activities. The chapters with distinguished activities are recognized by the Chapter Achievement Award (CAA). One chapter is selected from each of the four ComSoc Regions for the CAA, and the best performing chapter among those four CAA recipients is additionally recognized as the Chapter of the Year. Distinguished Lecturers/Speakers Programs: As a service to local chapters, ComSoc operates Distinguished Lecturer Tours (DLT) and individual Distinguished Speakers programs. These programs are arranged in response to requests from one or more chapter chairs, regional directors, or individual members. Usually up to five lecture tours are supported in each of the four regions every year. Distinguished Lecturers (DL) are selected by the DL Selection Committee, chaired by the Vice Chair of the Technical Activities Council (TAC). DLs are encouraged to network and exchange ideas with local chapter members while making technical lectures during DLT visits. Student Travel Grants (STG): ComSoc provides Student Travel Grants to help IEEE Student Members attend major ComSoc conferences. The number of grants is limited in number and offered to students who meet the eligibility requirement. Some conferences have additional sources for travel grants, e.g.; INFOCOM has the NSF program which supports travel for students studying at a US college or university. For more information on the STG program, please visit http://www.comsoc.org/about/documents/pp/4.1.5. Young Professionals Program: Within the framework of the IEEE GOLD program, ComSoc’s Young Professionals Program is designed to engage and support young member initiatives and leadership development. In support of this program, a Young Professional Web portal (YPW) was created and linked to ComSoc’s website, which focuses on career, education, on-line networking and engaging activities of members of interest. Please visit http://committees.comsoc.org/ypc/ for more information and participation. Industry Outreach Programs: The rapid transformation of the communications industry in the world’s developing as well as developed countries and their quest for technical excellence are creating an ever-increasing demand for networking and exchange of ideas and technologies. In response to such demands ComSoc proactively sets a framework to engage and support communications industry initiatives in these economies. The Society’s Corporate Patron Program and Industry Now Program are two representative programs that were developed to support these efforts. Corporate Patron Program (CPP): CPP is a package of ComSoc products and services that are specifically customized and bundled to meet the needs of each participant company. The program provides an opportunity for industry leaders to reach out to ComSoc’s influential members through exposure across the ComSoc web site, publications, and conferences. It enables companies to leverage ComSoc products and services through a package of customized discounts.
8
Communications IEEE
Industry Now Program (INP): INP is designed to promote industry participation in ComSoc activities around the world by offering companies and their employees the opportunity to use the values that ComSoc creates by working with professionals around the world. It is specifically tailored for industry organizations in the fast developing regions of the world. The program offers the option of customizing packages to address both geographic and company-specific needs, included in special membership packages. Industry Day: Recently, to jump-start IEEE’s interactions with global industry, ComSoc, in partnership with the IEEE and MGAB, organized an Industry Day event in Bangalore, India (http://ieee-industry.org/india/) with participation of over 400 delegates from 150 companies, government and academic
Opening session of India Industry Day. institutions (photo below). Industry Services Program (ISP): ISP, a new program, is being developed to meet the special needs of industry members in different countries. This program will offer mentoring service for industry members throughout their careers; information filtering services for the practicing engineer; unbiased product information summaries; business guidance services; and job search assistance services. Those component services will be designed, tested, and rolled out in the near future.
SISTER SOCIETIES WORLD-WIDE The ComSoc Sister Society program was one of the first tangible efforts by our Society to reach out to the global community of professionals who share our interests and values. Through establishing such relationships, we are able to generate interest in our activities, and in some cases formally collaborate on conference and publication activities (e.g., the Journal of Communications and Networks with our Sister Society in Korea, and the China Communications Magazine with our Sister Society in China). In many parts of the world, Sister Society colleagues are also active in ComSoc Chapters. This provides an opportunity for even more synergy across the professional communities. The “ComSoc/KICS Exemplary Global Service Award” jointly created by ComSoc and KICS (Korea Information and Communications Award) to recognize those who contributed to the exchange of technology and networking globally is in itself an excellent example of inter-Society collaboration and synergy. ComSoc is now connected to 30 Societies around the world and has a significant international footprint (see photo at the top of the next page). There are numerous reasons for having established these links, including the possibility of creating a much larger community (globally our Sister Society members exceed 500,000 engineers, scientists, and other professionals) and the fact that each Society, in a way, brings a unique perspective that enriches the ComSoc ecosystem.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
THE PRESIDENT’S PAGE
World map of ComSoc’s Sister Societies.
Attendees at the AP-RCCC in Kyoto, Japan.
Over the years we have developed important programs for cooperation with Sister Societies. For example, we jointly organize conferences, where ComSoc is ready to provide keynote speakers and tutorials by leveraging its member base. This increases attendance by bringing international participation to our Sister Society conferences and bringing local attendance to ComSoc’s conferences. In addition, these provide forums for interaction and networking among those with broader interests. Through ComSoc’s services, we advertise the activities of the Sister Societies in IEEE Communications Magazine and on the ComSoc web site. Papers published in IEEE Communications Magazine can be reprinted in Sister Society publications and, in some cases (e.g., two Sister Societies in China), with translation and republication in local languages. Recently, the ComSoc BoG took three significant steps to advance the Society’s relationships with its Sister Societies: (a) provide a link to ComSoc chapters in the geographical area of a Sister Society by appointing a local ComSoc representative in the Sister Society; (b) extend Sister Societies’ participation in areas such as Africa and Australia where the membership coverage is limited; and (c) insert in renewal agreements a specific goal that has to be achieved during the renewal period to make our relationship measurable and more effective in meeting cross-Society needs. We expect that these advancements will all be implemented by the end of this year.
In 1998, APR was the first ComSoc Region to create the Asia Pacific Young Researcher Award. It is presented to selected young researchers who have demonstrated distinguished performances and very active participation in ComSoc publications and conferences over the last three years. APR hosted an Asia Pacific-Regional Chapter Chairs Congress (AP-RCCC) in Kyoto in June, co-located with ICC 2011 (see photo above).
COMSOC ASIA PACIFIC REGION (APR) APR has 41 ComSoc chapters and has over 10,000 IEEE members. It covers a geographical area stretching from South Korea and Japan in the north-east to New Zealand in the south, and Pakistan in the west, and has diverse and dynamic cultural backgrounds. APR has a very well organized operational structure centered around itsAsia Pacific Board (APB), and a large number of active volunteers. The APB operational structure includes a Director, two Vice Directors, and five Committees, namely: Technical Affairs Committee (TAC), Meetings and Conferences Committee (MCC), Information Services Committee (ISC), Membership Development Committee (MDC), and Chapters Coordination Committee (CCC). This APB structure, closely supported by ComSoc’s Singapore Office, very effectively propels APR activities. APR has been very energetic in hosting major international conferences and has organized a number of annual regional conferences. Being dynamic and the fastest developing Region, it has shown significant growth in membership, chapters, conferences, and DLT/DSP visits. To keep its members informed, the region regularly publishes a newsletter, AP Newsletter, and contributes to ComSoc’s Global Communications Newsletter (GCN).
COMSOC EUROPE, MIDDLE EAST, AND AFRICA REGION (EMEAR) Geographically, EMEAR covers the largest area of four ComSoc Regions, extending from the Azores to Vladivostok, and from Aalborg to Cape Town. The Region has more than 10,000 ComSoc members. The volunteers in the Region are active in every aspect of ComSoc interest, including various member services, highly effective chapters, and successful, high quality conferences. EMEAR includes 49 chapters, spanning a very wide geographical area. The chapters are providing support and services to its members through regular meetings, topical lectures, DLT/DSP visits, and industry engagements. The number of chapters has been steadily increasing, with the addition of two to three new chapters each year. In 2010 the Chapter Achievement Award of EMEAR was presented to the Russia Siberia (Tomsk) Chapter. In 2010 EMEAR hosted a number of conferences on communications, including the IEEE International Conference on Communications (ICC 2010) in Cape Town, South Africa. This conference was successful in providing participants with a unique cultural experience and in connecting with industry and academic institutions and student organizations in the Region. In 2010 EMEAR established a ComSoc Young Researcher Award to recognize talented young members who have made significant original research contributions and actively participated in ComSoc publication and conference activities over the last three years. Dr. Joao Barros, at the University of Porto, Portugal, was selected as the first recipient of the award.
COMSOC LATIN AMERICA REGION (LAR) LAR is currently in a growth phase for its activities and membership. ComSoc is well represented throughout the Latin American continent in all countries from Mexico to Argentina with 24 chapters. Regular conferences at a national level are organized by its chapters in partnership with IEEE Sections, local universities and industry. IEEE LATINCOM is an annual communications conference organized by ComSoc’s Latin America Board (LAB). It
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
9
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
THE PRESIDENT’S PAGE As ComSoc continues its globalization reach to form the “Global ComSoc” village, NAR will continue to contribute as the Region with the longest history and a wealth of expertise.
FUTURE DIRECTIONS
Volunteer leaders at LA-RCC 2011, held in Cancun. is the largest communications conference in LAR and features significant international participation. LATINCOM not only serves as a major forum for technical interchange in NAR but also offers an opportunity for networking and sharing of cultural experiences. LATINCOM 2011 was held in Bogota, Columbia and will be held in Belem, Brasil in 2012. This year LAR hosted the LA Regional Chapter Chairs Congress (RCCC) in Cancun, Mexico, on March 30-31, colocated with WCNC (see photo above) and the BoG’s Operating Committee meeting. RCCC provided an opportunity to re-establish the relationship among chapters in LAR, and to conveniently meet with ComSoc’s leadership and staff. During the congress, an LAR strategic plan and a workable action plan were developed.
COMSOC NORTH AMERICA REGION (NAR) NAR is the largest ComSoc Region with almost 20,000 members in 93 chapters in the U.S. and Canada, which is about 40% of overall ComSoc organization. The size of chapters varies broadly from the Santa Clara Valley Chapter with 1,100 members to small chapters with 30-40 members. The leadership body of NAR, the North America Board (NAB), was created in 2008 with 10 members representing all NAR areas. Each board member plays two roles: representation of his/her own chapter, and leadership responsibilities to support the whole NAR, which includes all activities related to student, DLT/DSP, industry relations, membership development, information services, and GOLD programs. For years, the DLT has been a most demanding program in NAR. It has successfully stimulated cooperation among multiple chapters and helped membership development. In recent years, an increasing number of lecturers have been coming to NAR from other Regions. Such a trend directly serves ComSoc’s globalization goal and also serves our local members by providing additional value through cultural diversity. In the spirit of our Golden Triangle initiative, many of our regional distinguished lecturers and ComSoc officers have been delivering lectures in different Regions around the globe. NAR uses the North America-Regional Chapters Chairs Congress (NA-RCCC) as a forum for exchange among all constituent chapters.
10
Communications IEEE
We are proud of the ComSoc globalization accomplishments that have progressed far ahead of other Societies and have been leading the way within the IEEE. The geographic stretch of ComSoc’s membership and its world-wide activities is certainly making ComSoc a global Society, truly deserving the name “Global ComSoc.” The contributions of our members to ComSoc’s publications and conferences as well as their participation in technical and regional activities of APR and EMEAR, and the rapidly growing efforts by LAR, have already substantially improved the global balance across regions. In parallel, ComSoc’s global coverage has been steadily expanding, reaching 207 regional chapters and collaborating with 30 Sister Societies. In response to that growth we have diversified our membership programs so that they can weave the global networks and activities of our members worldwide. Globalization may be defined by two levels. With globalization at the membership level, or Level-1 Globalization, reached, we are now working to step up to the next level, Level-2 Globalization, where globalization is realized on the Society’s leadership team, including the Board of Governors the Society’s various councils, boards, and committees. Achieving balanced regional representation for ComSoc’s leadership team is very important for assuring ComSoc’s growth in the global era, through enabling ComSoc to operate in a truly global manner through making good use of global cultures and values. ComSoc has already succeeded in establishing a basic level of regional balance by appointing Regional Directors to its Board of Governors. The next step is to move forward toward a balanced representation of its Members-at-Large (MaL), who are voting members of the ComSoc BoG (in addition to other elected officers, such as the President and Vice Presidents). In practice, it can be realized by electing three MaLs in APR, three MaLs in EMEAR, one MaL in LAR, and five MaL in NAR, in approximate proportion to the number of ComSoc members in those regions. Since 2009, ComSoc’s BoG has been working toward this through its ad hoc Nomination and Election Process Committee. We anticipate Level-2 Globalization will be a reality within the next few years. The 21st century has opened a new era in our ongoing quest for a better and more humane world. While physical divides still exist in some parts of the world, disparity in economic, social and political life has a chance, for the first time in the history of mankind, to be patched through “communications.” Global communications in the digital convergence era will offer an affordable means to communicate among all different sectors of humanity, regardless of where people live. Leveraging our technical strength, creativity, and collaboration among our “Global ComSoc” membership, we now have the opportunity to capitalize on this movement and serve our members and humanity at large. We invite you to join us in meeting this most important challenge.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
TM
This presentation provides the latest technical developments and business/market potential for what promises to be the most prominent wireless technology ever to be deployed.
IEEE
BEMaGS F
____________________________________________________
The trend towards All-IP networks and increased demand for data services prompted the 3GPP to assess the implications for UMTS and High Speed Packet Access (HSPA) technologies. Even though these technologies will be highly competitive in the near term, to ensure competitiveness over a longer time frame, the 3GPP realized the need for a long-term evolution of the radio-access technology and an optimization of the packet core network. Research, development and standardization in these areas have led to the definition of Evolved UTRAN (E-UTRAN) and Evolved Packet Core (EPC) specifications.
Communications
A
FREE ACC CESS
sponso ored by
Brought to you by
For other sponso sorr opportunities, please con onta tact Eric Levine, As sso soci ciate Publisher Phone: 212-705-8920, E-mail:
[email protected] _____________
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
CONFERENCE REPORT IEEE ICC 2011 EXPLORES GLOBAL ADVANCE OF WIRELESS SYSTEM AND NETWORKING COMMUNICATIONS IN KYOTO, JAPAN Themed the “Source of Innovation: Back to the Origin,” the IEEE International Communications Conference (ICC 2011) recently concluded its latest annual event with over 1,800 international experts participating in the presentation of nearly 1,200 technical papers, symposia, tutorials and workshops dedicated to the ongoing advance of next-generation voice, data,and multimedia services, and theory and technologies for wireless and optical communications. Held in the fabled city of Kyoto, which reigned as Japan’s capitol for 1200 years and is recognized throughout the world as the cultural heart of Japan, IEEE ICC 2011 began on Sunday, June 5 with the first of two full days of workshops and tutorials detailing the latest research in topics ranging from “Visible Light Communications,” “Practical Solutions for Cognitive Radio,” and “Next Generation Broadband Access” to “Heterogeneous Networks (HETnet),” “Smart Grid Communications” and “Game Theory and Resource Allocation for Future Wireless Systems.” The following morning, the conference’s comprehensive symposium and business forum officially commenced with introductions from IEEE ICC 2011 General Chair Noritaka Uji, Executive Chair Koichi Asatani, IEEE ComSoc President Byeong Gi Lee and TPC Chair and IEICE-CS President Kazuo Hagimoto. In addition to exploring the event’s goal of enhancing life through the proliferation of next wave information communications, each speaker expressed their individual admiration for the Japanese people, who were so deeply affected by the recent tragedies befalling the nation just a few months ago. As a result, the entire forum joined in a moment of silence to honor the loved ones lost during the devastating March 11 event. Immediately afterwards, Ryuji Yamada, President & CEO of NTT DOCOMO, INC. detailed his company’s “Actions for Growth” in the Japanese marketplace as well as its “Response to the Great East Japan Earthquake” in the conference’s first keynote address. This included moving rapidly and steadily to restore the capabilities of 4,900 base stations to pre-disaster levels by April 30 and placing a high-priority on the use of radio signal transmission systems, satellite circuits and highperformance antennas to supply service to the many areas surrounding the Fukushima Daiichi Nuclear Plant. As for the future, Yamada spoke about the growing proliferation of
Professor Maurizio Decina of the Politecnico di Milano in Milan, Italy, addressed his vision of "Future Networks & Services."
12
Communications IEEE
Ryuji Yamada, President & CEO of NTT DOCOMO, INC. detailed his company's "Actions for Growth" in the Japanese marketplace. smart phones in a Japanese marketplace that already “leads the world in the adoption of 3G mobile communications services” and his company’s “aim to sell six million units of smart phones in 2011,” which will be armed with various first-ofkind services including real-time translation capabilities. Following this presentation, Professor Maurizio Decina of the Politecnico di Milano in Milan, Italy, addressed his vision of “Future Networks & Services” in a world that currently includes 5.3 billion mobile subscriptions and trillions of dollars in revenues. In the future, Decina envisions a “flatter and much more densely interconnected Internet” comprised of mobile devices completing the “billions of simultaneous transactions” needed to mesh up personal data, preferences and devices in real-world time. This includes the introduction of cheap, easy and convenient on-demand services that “knows you and what is around you,” “learns what you like,” “discovers things relevant to you” and “filters out the irrelevant.” At the end of these keynotes, attendees were then invited to attend any or all of the 1,000 technical paper presentations and business forums designed to explore the full array of information security, wireless networking, communication theory, signal processing and cognitive radio issues over the next three days. Among them were the hosting of nearly one dozen high-level executive panels that started with Business Forum Co-Chair Chi-Ming Chen moderating the opening discussion of “Open Innovation and Standardization Toward Next Generation Visual Networks” and “Dependable Wireless Communications.” Tokumichi Murakami of the Mitsubishi Electric Corp. in Japan began this session with his presentation on “High Efficiency Video Coding for Smart Phone to Super High Vision.” In addition to declaring the “2010s as the era of the smart phone, which will accelerate the personal use of visual contents and give rise to a new wave of content explosion,” Murakami spoke about the need and effort to develop new video coding standards targeting compression performances in order to realize the full-capabilities of future services such as SHV/UHD and Mobile HD. Afterwards, Professor Ryuji Kohno of the Medical ICT Center at Yokohama National University, continued the discussion of future technologies by speaking about “Future M2M for Medicine, Vehicle, Robot, Energy and others.” His vision also detailed the development of “safe and secure
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
CONFERENCE REPORT infrastructures based on advanced ICT” that support “the intelligent traffic control of energy, money, vehicles and medical flows.” As an example, Prof. Kohno cited the ongoing development of Body Area Networks (BAN) as a new method for improving efficiencies, reducing regional gaps and offering real-time medical care through the tele-metering of vital signs and tele-control of medical equipment and devices. Throughout Monday afternoon, IEEE ICC 2011 continued its high-level agenda with several General Industry Business Forums dedicated to next generation wireless advances and applications. Moderator Stan McClellan of Texas State University initiated the panel on “Standards, Technologies and Platforms for Emerging Smart Grid Deployments,” which highlighted various perspectives on grid stability and the numerous challenges facing the modern use of Smart Grid technologies ranging from interoperability to the development of competing standards and complex architectural paradigms. Following this forum, Stefano Bregni of the Politecnico di Milano in Italy led a second panel on “Next Generation Access Networks (NGAN): Ultrabroadband Infrastructures and Services.” Through this seminar, panelists examined the deployment of optical fiber access technologies in Europe and Asia with reference to the adoption of specific network architectures, services and regulations. At the end of day, Yoshihito Sakurai of Hitachi Ltd, in Japan then moderated the session entitled “Communication Technology for Smart Grid & Its Standardization,” which explored the newest cases, requirements and architectures for implementing smart gird communications that provide realtime information to international markets, while greatly reducing costs and saving energy on multiple levels. On Tuesday morning, IEEE ICC 2011 began with a commemorative ceremony led by IEEE ICC 2011 Executive Chair Prof. Koichi Asatani, who welcomed numerous Japanese and fellow IEEE dignitaries to the day’s events. This included Vice Governor of Kyoto Prefecture Shuichi Yamauchi, Vice Mayor of Kyoto City Fumihiko Yuki, the President of the Science Council of Japan Prof. Ichiro Kanazawa, the President of IEICE Communications Society Kazuo Haigmoto, the President of IEEE Communications Society Prof. Byeong Gi Lee, His Imperial Highness Prince Akishino and the General Chair of IEEE ICC 2011 Noritaka Uji. Following these introductions, Mr. Uji, Prof. Kanazawa and His Imperial Highness addressed the forum by offering their special thanks to all IEEE ICC 2011 participants for their attendance and expressing their deepest sympathies to the many individuals and families that recently suffered devastating losses. This also included strongly highlighting conference objectives and the overall goal of enriching lives through the deployment of the latest ICT applications, services and innovations. Afterwards, Prof. Asatani read a message from the Prime Minister of Japan who also welcomed all and again profoundly cited the advance of electronic and information communications worldwide as a key method for significantly reducing the horrific affects of future and potential global disasters. Dr. Toshitaka Tsuda of Fujitsu Laboratories Limited then opened the day’s educational agenda by offering his gratitude to all the overseas attendees, who traveled to the beautiful and peaceful city of Kyoto for IEEE ICC 2011. Following these remarks, Dr. Tsuda spoke about the new “ICT Paradigm Shift and the Trend of Communications Technology.” This includes moving toward “Human Centric Systems” and “Human Centric Intelligent Societies” that are specificallydesigned to transform large amounts of data acquired through sensor networks into knowledge that initiates social and business changes. According to Dr. Tsuda, this shift is currently being realized by the ongoing adoption of innovative systems
that place individuals and their surrounding business environments at the center of learning networks that constantly evolve to new requirements. Immediately after the keynote address of Dr. Tsuda, representatives of IEEE ComSoc and IEEE ICC 2011 highlighted the morning’s proceedings with the presentation of numerous industry, association and conference awards signifying the outstanding achievements and dedication to excellence of the international recipients. Senior IEEE and IEEE ComSoc representatives offered the Marconi Prize Paper Award to Li Ping Qian, Angela Yingjun Zhang & Jianwei Huang; Stephen O. Rice Award to Shi Jin, Matthew R. McKay, Xiqi Gao & Iain B. Collings; Fred Ellersick Prize to Dusit Niyato, Ekram Hossain & Zhu Han; Heinrich Hertz Award to Illsoo Sohn & Jeffrey G. Andrews; Leonard Abraham Prize to Watcharapan Suwansantisuk, Marco Chiani & Moe Z. Win; William Bennett Prize to Murali Kodialam, T. V. Lakshman, James B. Orlin & Sudipta Sengupta; Outstanding Paper on New Communication Topics to Maria Gorlatova, Peter Kinget, Ioannis (John) Kymissis, Dan Rubenstein, Xiaodong Wang & Gil Zussman; Best Tutorial Paper Award to Steven Gringeri, Bert Basch, Vishnu Shukla, Roman Egorov & Tiejun J. Xia; and the Journal of Communications & Networks (JCN) Best Paper Award offered by the KICS society and cosponsored by ComSoc to Changhun Bae and Wayne E. Stark. Others honored throughout the Tuesday’s ceremony included Suk-Chae Lee, who received the Distinguished Industry Leader Award; H. Vincent Poor, who was provided the Eric E. Sumner Award; Moe Win, who was named the Kiyo Tomiyasu Award winner; Andreas F. Molisch, Larry J. Greenstein & Mansoor Shafi, who were all granted the Donald G. Fink Prize Paper Award; and Mounir Hamdi, Yasutaka Ogawa, Yunghsiang S. Han, Marco Chiani, Kwang Bok Lee, Kiho Kim, John Sadowsky & Robert Heath, who were all named 2011 IEEE Fellows. Furthermore, the Communications Society/Information Theory Joint Paper Award was presented to two separate entries. The first was provided to Matthieu Bloch, Joao Barros, Miguel R. D. Rodrigues, and Steven W. McLaughlin for the paper titled “Wireless Information-Theoretic Security, while the second was given to Giuseppe Caire, Nihar Jindal, Mari Kobayashi & Niranjay Ravindran for their paper on “Multiuser MIMO Achievabole Rates with Downlink Training and Channel State Feedback.” The awards presentations then concluded with the IEEE ICC 2011 Awards for Best Papers. These included: •Prathapasinghe Dharmawansa & Matthew R. McKay of Hong Kong University of Science and Technology and Peter J. Smith of the University of Canterbury, New Zealand for their presentation on Analysis of the Level Crossing Rates for Ordered Random Processes in the Communication Theory category •Ya-Feng Liu & Yu-Hong Dai of the Chinese Academy of Sciences and Zhi-Quan Luo of the University of Minnesota for their submission on Max-Min Fairness Linear Transceiver Design for a Multi-User MIMO Interference Channel within the conference’s Signal Processing for Communications section •Shuping Gong & Husheng Li of the University of Tennessee, Lifeng Lai of the University of Arkansas and Robert C. Qiu of the Tennessee Technological University for their entry titled Decoding the ‘Nature Encoded’ Messages for Distributed Energy Generation Control in Microgrid in the Wirelesss Communications category •Mingwei Wu & Pooi-Yuen Kam of the National University of Singapore for their submission on ARQ with PacketError-Outage-Probability QoS Measure under the Wireless Communications heading
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
13
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
CONFERENCE REPORT •Xiangyun Zhou & Are Hjørungnes of the University of Oslo, Norway and Dusit Niyato of the Nanyang Technological University in Singapore for their submission titled How Much Training is Needed Against Smart Jamming? in the Wireless Communications area •Dejun Yang, Xi Fang & Guoliang Xue of Arizona State University for their paper on OPRA: Optimal Relay Assignment for Capacity Maximization in Cooperative Networks in the area of Wireless Networking •Chengyi Gao, Hakki Cankaya, Ankitkumar Patel & Jason P. Jue of the University of Texas, Dallas and Xi Wang, Qiong Zhang, Paparao Palacharla & Motoyoshi Sekiya of Fujitsu Laboratories of America, Inc., USA for their submission on Survivable Impairment-aware Traffic Grooming and Regenerator Placement with Dedicated Connection Level Protection in the Optical Networks and Systems section •Hossam Sharara of the University of Maryland and Cedric Westphal, Svetlana Radosavac & Ulas Can Kozat of DoCoMo Labs USA for the entry titled Utilizing Social Influence in Content Distribution Networks in the Next Generation Networking and Internet category •Jesus Alonso-Zarate & Christos Verikoukis of the Centre Tecnologic de Telecomunicacions de Catalunya, Spain and Luis Alonso of the Universitat Politecnica de Catalunya, Spain and Eirini Stavrou, Adamantia Stamou & Pantelis Angelidis of the University of Western Macedonia, Greece for their entry on Energy-Efficiency Evaluation of a Medium Access Control Protocol for Cooperative ARQ in the area of Communications QoS, Reliability and Modeling •Jalel Ben-othman & Bashir Yahya of the Université de Versailles, France and Lynda Mokdad of the Université de Paris 12, France for their submission on An Energy Efficient Priority-based QoS MAC Protocol for Wireless Sensor Networks in the Ad Hoc, Sensor and Mesh Networking •Jin Tang & Yu Cheng of the Illinois Institute of Technology, USA for the paper titled Quick Detection of Stealthy SIP Flooding Attacks in VoIP Networks within the conference’s Communication and Information System Security category Following the morning’s keynote and commemorative ceremony, Business Forum Co-Chair Chi-Ming Chen moderated the first of several Business Forums dedicated to the development of the latest broadband technologies and their rapid worldwide deploy. In his address on “Telecom Transformation and Challenges,” Shyue-Ching Lu, CEO of Chungwa Telecom in Taiwan opened the session by stating how networked readiness has become a significant index for rating a nation’s competence. He then noted the steady paradigm shifts in all technological areas including e-government, smart transportation, i-education, healthcare and disaster prevention, which are currently doubling the amount of global data traffic every two years, while only increasing network energy efficiencies by 10 to 20 percent annually. As a result, Shyue-Ching Lu prominently spoke about the need for initiatives, such as those offered by the Green Touch Consortium, that are actively promoting the development and implementation of architectures and technologies that will yield a 1,000-fold improvement in network efficiencies over the next five years. This includes the digital convergence of smart devices and cloud services that provide centralized monitoring and management for power, air conditioning, lighting and pumping systems in facilities ranging from commercial buildings and factories to hospitals and schools. Afterwards, Toshitaka Tsuda of Fujitsu Laboratories Limited, Japan furthered the comments made during his earlier address by speaking about “R&D for Green and Human Centric Networks” as well as the requirements necessary to accommodate the increasing demand for bandwidth and green communications services. He also explored the rapid adoption
14
Communications IEEE
and innovation of network technologies that are proactively transforming business processes and creating vast pools of knowledge that support human wide geometric areas and spontaneous smart transmissions. Ibrahim Gedeon, IEEE ICC 2012 General Chair & CTO of Telus Canada, then concluded the morning business session by addressing the need to continually transform businesses through leveraged technology convergence practices. According to Gedeon, what is needed is a holistic view of convergence based on end-to-end business models that improve end-user experiences. This is because “one thing is clear” and that is end-users care most about the value of their experience and the amount they have to pay for it. After a brief afternoon break on Tuesday afternoon, Business Forum Co-Chair Tetsuya Yokotani began his third Business Forum of the conference by introducing Yoshiharu Shimatani of KDDI, Japan to session attendees. In his presentation, Shimatani spoke at-length about “KDDI’s Vision of Innovative, Seamless and Eco-friendly ICT Platforms” and the need to effectively alleviate traffic congestion through the coordination of seamless multi-networks that support robust and sophisticated anytime, at anywhere applications. These methods entailed alleviating the rapid rise of traffic congestion through the development of innovative ICT platforms and cloud services that drive the newest e-health, e-education, e-environment, e-disaster prevention infrastructures. Tetsuya Yuge of Softbank Mobile Corporation, Japan, then followed this presentation by highlighting his company’s “Wireless Broadband Strategy for Growing Mobile Data Traffic” and accommodating the prevalent use of smart phones in the Japanese marketplace and the societal benefits to its businesses and culture. Young Ky Kim of Samsung Electronics, Korea also leveraged the theme by exploring the “Crossroads in the Middle of Mobile Big Bang Era” and the mobile industry of the future.” Throughout the rest of Tuesday and Wednesday, IEEE ICC 2011 proceeded with the presentation of numerous other General Business Forums provided by the representatives of leading worldwide corporations and academic institutions such as Hitachi, Mitsubishi, NTT, Create-Net, DOCOMO, University of Zurich, Politecnico di Milano, University of Geneva, UC Berkley and Auckland University of Technology in New Zealand. During these sessions, experts from each facility readily shared their views and research in areas that included “Scientific Wireless Sensor Network Testbeds: Growth & Impact,” “Green ICT,” “Business Showstoppers of Cognitve Radio Technologies,” “Business Strategies of sustainable Growth on Broadband Telecommunication Markets,” “eHealth Support in the Future Internet.” On Thursday, June 9, IEEE ICC 2011 then concluded the conference’s extensive five-day agenda with a second day of tutorials and workshops dedicated to key communications like “MIMO Detection for Emerging Wireless Standards,” “Wireless Without Batteries,” Beyond IMT-advanced: The Next 10 Years,” “Green Communications & Energy Efficiency,” “Advances in Mobile Networking” and “Embedding the Real World in the Future Internet.” As for IEEE ICC 2012, planning has already begun for the next event to be held June 10 – 15 in Ottawa, Canada. Titled “CONNECT • COMMUNICATE • COLLABORATE,” original papers detailing the latest advances in wide ranging communications are currently being accepted with a deadline date of September 6, 2011. For more information in this premier annual conference, including “Call for Papers” details please visit http://www.ieee-icc.org/2012. All interested parties are also invited to use the conference’s Facebook and Twitter pages to share thoughts about IEEE ICC experiences and upcoming attendance and scheduling plans with peers and colleagues.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
TM
A
BEMaGS F
____________________________________________________
This presentation from the recent IEEE Communications’ Conference on Consumer Communications& Networks discusses emerging technology for next-generation television and video applications and services. International standards (High efficiency video coding, stereoscopic 3D) for deployment of new servicesare covered, along with IPTV and dynamic adaptive streaming on HTTP for internet video delivery, and 1080p50/60 and ultra high-resolution television.
FREE ACCESS SPONSORED BY
Brought to you by
For other sponsor opportunities, please contact Eric Levine, Associate Publisher Phone: 212-705-8920, E-mail: ______________
[email protected]
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
CONFERENCE CALENDAR z IEEE EDOC 2011 - 15th IEEE Int’l. Enterprise Distributed Object Computing Conference, 31 Aug.-2 Sept.
2011 JULY z IEEE HPSR 2011 - 12th IEEE Int’l. Conference on High Performance Switching and Routing, 4-6 July Cartagena, Spain. http://www.ieee-hpsr.org/
• OECC 2011 - 16th Opto-Electronics and Communications Conference, 4-8 July Kaoshlung, Taiwan. http://www.oecc2011.org/
z IEEE ICME 2011 - 2011 IEEE Int’l. Conference on Multimedia and Expo, 11-15 July Barcelona, Spain. http://www.icme2011.org/
AUGUST • ICCCN 2011 - Int’l. Conference on Computer Communications and Networks 2011, 1-4 Aug. Maui, Hawaii. http://www.icccn.org/ICCCN11/
z ATC 2011 - 2011 Int’l. Conference on Advanced Technologies for Communications, 3-5 Aug. Da Nang City, Vietnam. http://rev-conf.org/
• ICADIWT 2011 - 4th Int’l. Conference on the Applications of Digital Information and Web Technologies, 46 Aug. Stevens Point, WI. http://www.dirf.org/DIWT/
• GIIS 2011 - Global Information Infrastructure Symposium, 4-7 Aug. Da Nang City, Vietnam. http://www.giis2011.org/GIIS2011/index.htm
• ITST 2011 - 11th Int’l. Conference on ITS Telecommunications, 23-25 Aug. St. Petersburg, Russia http://www.itst2011.org/
z IEEE P2P 2011 - IEEE Int’l. Conference on Peer-to-Peer Computing, 31 Aug.-2 Sept. Tokyo, Japan. http://p2p11.org/
Helsinki, Finland. http://edoc2011.cs.helsinki.fi/edoc2011/
Palermo, Italy. http://www.fitce2011.org/
IEEE
• WPMC 2011 - 14th Int’l. Symposium on Wireless Personal Multimedia Communications, 3-7 Oct. Brest, France. http://www.wpmc2011.org/
SEPTEMBER • ITC 23 2011 - 2011 Int’l. Teletraffic Congress, 6-8 Sept.
• ICIN 2011 - 2011 15th Int’l. Conference on Intelligence in Next Gneration Networks, 4-7 Oct.
San Francisco, CA. http://www.itc-conference.org/2011
Berlin, Germany. http://www.icin.biz/
z IEEE PIMRC 2011 - 22nd IEEE Int’l. Symposium on Personal, Indoor and Mobile Radio Communications, 11-14 Sept.
• LANOMS 2011 - 7th Latin American Network Operations and Management Symposium, 10-11 Oct.
Toronto, Canada. http://www.ieee-pimrc.org/2011/
• ICUWB 2011 - 2011 IEEE Int’l. Conference on Ultra-Wideband, 14-16 Sept. Bologna, Italy. http://www.icuwb2011.org/
• ICCCT 2011 - 2nd Int’l. Conference on Computer and Communication Technology, 15-17 Sept. Allahabad, India. http://www.mnnit.ac.in/iccct2011/
• SoftCOM 2011 - Int’l. Conference on Software, Telecommunications and Computer Networks, 15-17 Sept. Split, Croatia. http://marjan.fesb.hr/SoftCOM/2011/index.ht ml _
z IEEE DSA 2011 - IEEE Dynamic Spectrum Access Workshop, 19 Sept.
Quito, Ecuador. http://www.lanoms.org/2011/
z IEEE CCW 2011 - 2011 IEEE Annual Computer Communications Workshop, 10-12 Oct. Hyannis, MA. http://committees.comsoc.org/tccc/ccw/2011/
• DRCN 2011 - 8th Int’l. Workshop on Design of Reliable Communication Networks, 10-12 Oct. Krakow, Poland. http://www.drcn2011.net/index.html
• IEEE ICWITS 2011 - 2011 IEEE Int’l. Conference on Wireless Information Technology and Systems, 10-13 Oct. Beijing, China. http://icwits.ee.tsinghua.edu.cn/index.html
z IEEE SmartGridComm 2011 - IEEE Int’l. Conference on Smart Grid Communications, 17-20 Oct.
http://www.ieee-dsa.org/IEEE_DSA_2011.html
Brussels, Belgium. http://www.ieee-smartgridcomm.org/2011/
• APNOMS 2011 - 13th Asia-Pacific Network Operations and Management Symposium, 21-23 Sept.
z IEEE LANMAN 2011 - 18th IEEE Workshop on Local & Metropolitan Area Networks, 20-21 Oct.
Taipei, Taiwan. http://apnoms2011.cht.com.tw/Home.html
Chapel Hill, North Carolina. http://www.ieee-lanman.org/
z IEEE GreenCom 2011 - Online Conference, 26-29 Sept.
• LATINCOM 2011 - IEEE Latin American Conference on Communications 2011 - 246-26 Oct.
Virtual. http://www.ieee-greencom.org/
before the listing. Individuals with information about upcoming conferences, calls for papers, meeting announcements, and meeting reports should send this information to: IEEE Communications Society, 3 Park Avenue, 17th Floor, New York, NY 10016; e-mail: ____________
[email protected]; fax: +1-212-705-8996. Items submitted for publication will be included on a space-available basis.
Communications
Seoul, Korea. http://www.ictc2011.org/main/
OCTOBER • FITCE 2011 - 50th FITCE Congress ICT: Bridging the Ever Shifting Digital Divide, 31 Aug.-3 Sept.
z Communications Society portfolio events are indicated with a diamond before the listing; • Communications Society technically co-sponsored conferences are indicated with a bullet
16
• ICTC 2011 - Int’l. Conference on ICT Convergence 2011, 28-30 Sept.
Belém, Pará, Brazil. http://www.ieee-latincom.ufpa.br/
• ITU WT 2011 - Technical Symposium at 40th ITU Telecom World 2011, 2427 Oct. Geneva, Switzerland. http://www.itu.int/WORLD2011/index.html
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
F
SOCIETY NEWS COMSOC 2011 ELECTION TAKE TIME TO VOTE Ballots were mailed and emails with election information were sent 31 May 2011 to all Higher Grade* IEEE Communications Society Members and Affiliates (excluding Students) whose memberships were effective prior to 1 May 2011. To cast your ballot electronically you will need your IEEE Web Account username/password--which is the same account information used to access IEEE online services such as renewing your membership, myIEEE, and Xplore. If you do not recall your web account information, you may go to: www.ieee.org/web/accounts to recover. You may also email ________________
[email protected] or call +1 800 678 4333 (USA/Canada) or +1 732 981 0060 (Worldwide). If you do not receive an email from _______
[email protected] _____________ on 31 May 2011 or a paper ballot by 30 June, but you feel your membership was valid before 1 May 2011, you may e-mail
[email protected] ___________________ or call +1 732 562 3904 to check your member status. (Provide your member number, full name, and address.)
Please note IEEE Policy (Section 14.1) below stating IEEE mailing lists should not be used for electioneering in connection with any office within the IEEE: IEEE membership mailing lists, whether obtained through IEEE Headquarters or through any IEEE organizational unit, may be used only in connection with normal IEEE sponsored activities and may be used only for such purposes as are permitted under the New York Not-For-Profit Corporation Law. They may not be used for electioneering in connection with any office within the IEEE, or for political purposes, or for commercial promotion, except as explicitly authorized … . See details at url http://www.ieee.org/web/aboutus/ whatis/policies/p14-1.html. _________________ Voting for this election closes 26 July 2011 at 4:00 p.m. EDT! Please vote! *Includes Graduate Student Members
Open Call from the IEEE Communications Society
A
re you enthusiastic? Have you performed quality reviews for technical periodicals? Demonstrated solid technical accomplishments? Have a reputation for high ethical standards and for reliability?
www .com
soc.
org/
edit
or
You may be ready ... The IEEE Communications Society is looking for volunteers who are interested in becoming part of a prestigious Communications Society editorial board. Duties include: A commitment to handle at least two manuscripts per month; arrange for three reviews or more in a timely fashion; and the ability to make firm and fair decisions. Qualifications: Subject matter expertise, editing experience, technical program committee experience; references, representative papers.
Apply at: www.comsoc.org/editor The decision to appoint an editor rests with the Editor-in-Chief of the journal/magazine. Please note that it will not be possible to send individual acknowledgments to all applicants.
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
17
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
NEW PRODUCTS EDITED BY ERIC LEVINE COMPLETE MIPI M-PHY TEST SUITE Agilent Technologies Inc. Agilent Technologies has introduced a comprehensive MIPI M-PHY test solution for mobile computing customers. The Agilent solution suite helps design engineers turn on, debug and validate all layers of their M-PHY devices, including physical and protocol layers, at speeds up to 5.8 Gb/sec. The Mobile Industry Processor Interface (MIPI) Alliance is finalizing the M-PHY specification to allow development of faster, more reliable highspeed interfaces for mobile devices. M-PHY technology supports a broad range of applications, including interfaces for monitors, cameras, audio and video equipment, memory, power management and communication between baseband and RFIC components. The Agilent solution consists of oscilloscopes, protocol analyzers and exercisers, and bit error-rate testers (BERTs) using custom M-PHY stimulus software. Each instrument comes with custom M-PHY-ready software to support design teams through the entire product design process. The Infiniium 90000 X-Series oscilloscope real-time bandwidth of up to 32 GHz with a 30-GHz probing system. With low noise and low jitter measurement floor performance, the scope ensures superior accuracy and is ideal for MIPI M-PHY transmitter conformance testing for speeds up to Gear 3. Using 90000 X-Series oscilloscopes gives engineers increased confidence in their MIPI M-PHY product performance and increases design margins. Accurate and automated MIPI MPHY receiver testing is supported by Agilent’s high-performance ParBERT 81250A for multi-lane testing and JBERT N4903B for single-lane testing. Although the MIPI M-PHY receiver test specifications have not been finalized yet, engineers can use these bit error-ratio testers for accurate M-PHY receiver tolerance testing in a pattern generator or full BERT configuration in conjunction with N5181/2A, E4438C and 81150A signal generators. http://www.agilent.com
REMCOM ANNOUNCES UPDATE TO XFDTD EM SIMULATION SOFTWARE Remcom Inc. Remcom is offering an updated version of its electromagnetic simulation software, XFdtd Release 7 (XF7), which includes several geometric modeling additions and new performance
18
Communications IEEE
efficiencies. The upgrade, which updates the software to Release 7.2, contains enhancements that simplify and speed overall usability: •Additional modeling capabilities that enable more precise control over cut geometries when sketching complex models. •Simplified sampling interval settings for Frequencies of Interest (DFT). •Simulations using averaged materials now exploit XStream GPU acceleration technology. •XFSolver intelligently records which XStream devices are being used for simulations, allowing multiple simultaneous XStream simulations on a single machine. •Performance of Broadband Far-Zone (previously known as Transient FarZone) computations have been dramatically improved. •Additional efficiencies in CAD import, particularly for large models with many assemblies. XF7 is available in both Pro and Bio-Pro versions. Both include XStream GPU acceleration, 32- or 64-bit analysis module, geometric modeler and postprocessor, shared memory multiprocessor (MPM) at eight cores, and a comprehensive variety of 3D CAD import modules. The Bio-Pro version also includes SAR capability and high fidelity human body meshes. http://www.remcom.com
BROADBAND QUADRATURE MODULATORS FOR CELLULAR INFRASTRUCTURE MARKET Skyworks Solutions, Inc. introduced three wideband quadrature modulators for cellular infrastructure and high performance radio link applications. Skyworks’ modulators are the latest additions to its wireless infrastructure portfolio and designed to support the world’s leading 3G and 4G base station providers. These new, fixed gain quadrature modulators deliver excellent phase accuracy and amplitude balance enabling high performance for a variety of multi-carrier communication systems. In addition, Skyworks’ new modulators have greater than 500 megahertz (MHz) 3dB modulation bandwidth, a low noise floor, and a wide operating frequency range that support multiband designs and network requirements. The SKY73077 (for 1500 to 2700 MHz), the SKY73078 (for 500 to 1500 MHz), and the SKY73092 (for 400 to 6000 MHz), quadrature modulators contain high linearity, excellent I/Q
phase accuracy and amplitude balance – making the devices ideal for use in high performance communication systems. The modulators accept two differential baseband inputs and a single-ended local oscillator, and generate a singleended RF output. http://www.skyworksinc.com
LTE PROTOCOL SIMULATOR GL Communications Inc. GL Communications has released the LTE Protocol Simulator, a software application for emulating LTE interfaces. LTE is an all IP infrastructure with service priority built in – audio and video are given priority. All necessities like IP address, authentication, and security are validated. Instant resources over RF (the air) and IP (internal network) are made available depending on what the user is attempting to do. Also, LTE is designed for compatibility with older 2G and 3G mobile systems. GL has released LTE Protocol Simulation for several interfaces – currently S1 – MME, and eGTP (S5/S8 and S11). Its Message Automation & Protocol Simulation (MAPS™ - LTE-S1 and LTE eGTP) is designed for LTE Testing. It can simulate eNodeB (also called Evolved NodeB), and MME (Mobility Management Entity) and other interfaces. MAPS™ - LTE-S1 can act either as eNodeB or as MME and simulate the other entity. It can generate any LTES1 messages in an automated, interactive, and scripted fashion. Possible applications include: -•Simulate up to 500 Smartphones (UEs) powering up and down -•Authenticate and confirm security procedures -•QoS requests for greater or lesser bandwidth -•Temporary addressing management for mobility and security LTE-S1 Testing Features •Simulates eNodeB and MME •Supports LTE control plane •Generates hundreds of UE Signaling •Handles Retransmissions •Generates and process S1/NAS valid and invalid messages •Supports message templates for both S1-AP and NAS message and customization of the field values •Allows defining variables for various fields of the selected message type •Ready-to-use scripts for quick testing •Supports customization of call flow and messages •Supports scripted call generation and automated call reception http://www.gl.com
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Global
Newsletter July 2011
4G Mobile Licenses Under Auction in Spain By F. Falcone Lanas and Paula Pérez Gómez, Public University of Navarra and University of Vigo, Spain The Ministry of Industry announced in February 2010 [1] a plan to collect between €1500 and €2000 million with the auction of new frequency bands for mobile telephony. The bid for a total of 310 MHz in different frequency bands will be done by a mixed formula: 90 percent through auctions and 10 percent by competition. The available spectrum space is free as a result of the rearrangement of the frequency portion related to the previous generation of mobile telephony (GSM) and the spectrum vacated by the switch-off of analog TV services. Of the amount raised, €800 million will go to arrange the frequencies now occupied by digital terrestrial television (DTT), the socalled digital dividend in the range from 470 to 862 MHz. Several management approaches toward spectrum allocation are possible, the auction being among the preferred ones [2]. The Minister Miguel Sebastian pointed out that the decision of a mixed procedure has the desire to add transparency to the process as well as the tax collection increase. The Telecommunications Market Commission (TMC) noted in his report on the rearrangement of the spectrum that this auction system is more transparent than the tender-based one used in 2000 and 2005. The bid for 310 MHz is the most important spectrum restructuring undertaken in Spain and thus facing similar processes in Europe. The Ministry of Industry explained that the system wants to prioritize the investment of operators in the sector for the benefit of citizens and society with a large and direct impact on employment and productivity. Thanks to this formula, it is expected to mobilize an investment of €1200 million and create 40,000 jobs. The Ministry of Industry already has the Electronic Auction Platform (EAP) software to be used by interested parties to bid in a process expected to begin by July 2010 and to last as long as bids continue to rise. Government has limited the access to frequencies by traditional mobile operators (Movistar, Vodafone, and Orange) by establishing an upper limit for any operator to grab a majority of the spectrum, according to the TMC indications, so as not to harm competition or lock out new entry rivals. More than 30 companies have shown interest in the consultation regarding the process, far more than telecommunications operators in the country. The use of this bidding system is being pioneered in Spain, although some companies have experience in auctions held in other countries, such as Telefonica, which participated in the auction held in Germany in 2010 via its filial O2 and paid €1380 million for a set of frequency blocks: two in 800 MHz, one in 2 GHz, and four in 2.6 GHz. With the allocation of additional spectrum to be held in the second quarter of 2011, the government follows the exam-
ple of other countries to promote competition and enable new services for fourth-generation mobile radio by increasing the space available by 70 percent. In this rearrangement the band of 800 MHz would be bid on in auction; in the 900 MHz band, the block of 5 MHz will be awarded by competition and with availability last year, whereas the two blocks of 5 and 4.8 MHz was to be auctioned this year but will be available in 2015. The band of 1800 MHz would also be awarded by competition to be available in 2011, while the new 2.6 GHz band would be bid at auction and also available last year. The industry spent months awaiting the completion of the process to undertake the reorganization of the radio spectrum that would allow telecom operators to have more resources to grow, mainly in mobile broadband services. With this spectral reorganization, the regulation that allows the use of these frequency bands for mobile operators is put in practice, so the band of 900 MHz which has been only available to GSM may be used for broadband services; and also incorporated new frequency bands, such as 800 MHz and 2.6GHz. These frequencies will allow better coverage with less investment, but would not be available until 2015. They will also ensure coverage for ultra-fast mobile broadband to 98 percent of the population, thus facilitating the achievement of the objectives of the Digital Agenda for Europe [3] by 2020 and so strongly contribute to reducing the digital divide. In 2010 in Spain radio frequencies generated a business of €22,000 million, and will go further to meet the demand for mobile broadband. In Spain there are 54.3 million mobile lines (116.3 lines per 100 inhabitants), 92.5 percent owned by Telefonica (TEF.MC), Vodafone (VOD.L), and Orange (FTE.PA). Telstra (TLSN.ST) has a market share of 3.9 percent, and 3.5 percent is held by a dozen mobile virtual network operators (who have no network). The explosion of data services via mobile phones due to the proliferation of new devices like smart phones, laptops, and tablets, which require mobile Internet connection, has made those frequencies the desired target for all companies who fear the collapse of their networks. These indicators turned the Spanish auction into a promising business for international investors. Telecommunications operators interested in radio frequency bands of 800 MHz, 900 MHz, and 2.6 GHz had until 7 June to submit applications and participate in the auction process, which began before 30 June 2011. Orange is committed to invest a total of €433 million in the competition for frequencies of 900 MHz, while Telstra has offered a total of €300 million for the bands of 900 and 1800 (Continued on Newsletter page 4)
Global Communications Newsletter • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
1 A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Fourth IEEE Lebanon Communications Workshop (IEEE LCW ’10) Attracts More than 250 Participants By Zaher Dawy, IEEE ComSoc Lebanon Chapter Chair The IEEE Communications Society Lebanon Chapter organized the Fourth IEEE Lebanon Communications Workshop 2010 (IEEE LCW ’10) on 18 December at the Faculty of Engineering, Lebanese University (LU). The general theme of the workshop was Communications Security with focus on state-of-the-art topics related to network and wireless security. The IEEE LCW workshop is the biggest telecom event that is organized at the national level on an annual basis. This year, it attracted around 275 participants including telecom engineers, telecom executives, university professors, and engineering students. The program started with an opening session, followed by two technical sessions. The IEEE Communications Society Lebanon Chapter chair, Dr. Zaher Dawy (Professor, American University of Beirut), opened with a welcoming speech, thankful for the strong participation and distinguished speakers from Lebanon and abroad. He pointed out that IEEE LCW ’10 is instrumental in achieving the IEEE ComSoc Lebanon Chapter’s objectives: to contribute to telecom advancement on a national level, to add to telecom awareness, and to strengthen ties between the academic and industrial worlds. He concluded by thanking the Workshop’s corporate supporters for their generous contributions: National Instruments Arabia, MTC Touch, Mada Communications, Data Consult, Alfa managed by Orascom Telecom, and Terranet. The opening session included a presentation by Dr. Imad Elhajj (Professor, American University of Beirut) about the IEEE Lebanon Section activities highlighting the professional
Attendees at IEEE LCW ’10. benefits of IEEE membership, a speech by Dr. Imad Hoballah (Acting Chairman and CEO, Lebanese Telecom Regulatory Authority) on the importance of cyber security awareness at the national level, and a welcome speech by Dr. Mohamed Zoaeter (Dean of the Faculty of Engineering, LU). The technical program included featured presentations by distinguished invited speakers from the United Nations Interregional Crime and Justice Institute (UNICRI) on Cyberwar and Information Warfare by Mr. Raoul Chiesa (Senior Advisor, Strategic Alliances and Cybercirme Issues); Cisco on Borderless Network Security by Mr. Joseph Hanna (Business (Continued on Newsletter page 4)
New European Research Initiative on Techno-Economic Regulatory Framework for Cognitive Radio/Software Defined Radio By Arturas Medeisis, Chair of COST-TERRA, Vilnius Gediminas Technical University, Lithuania A new joint research initiative has recently been set up in Europe within the framework of the European Cooperation for Science and Technology (COST) to address the issue of a techno-economic regulatory framework for cognitive radio/software defined radio (CR/SDR). The initiative, formally known as COST Action IC0905 TERRA (COSTTERRA), is organized as a kind of “think tank” with regular networking meetings. Its planned activities span to May 2014. At the time of writing, the COST-TERRA network included researchers from 19 European countries representing academia, industry and regulators. Recently, members from other parts of the world have started joining the action. The first non-European membership was by Communications Research Centre of Canada, now being followed by the Meraka institute of CSIR South Africa and a couple of US institutions that are considering joining as well. COST-TERRA also established institutional liaison with bodies like CEPT (association of European regulators), European Telecommunications Standards Institute (ETSI), IEEE DySPAN (former SCC41) and the Wireless Innovation Forum (former SDR Forum). The meetings of COST-TERRA present an excellent opportunity for researchers as well as various players in the field to come for lively brainstorming sessions on the subject of developing regulatory policies for cognitive radio. The most recent meeting took place in Lisbon, Portugal, on 19–21 January 2011, and was hosted by the Instituto de Telecomunicações. The meeting was attended by 38 participants, and consisted
2
Communications IEEE
Participants of the COST-TERRA meeting in Lisbon. of both regular sessions to present the latest research in the field as well as panel discussions dedicated to hot issues such as regulatory policy for TV white space devices and developments with the CR/SDR-related agenda item for the ITU World Radiocommunications Conference of 2012. The next meeting of COST-TERRA will take place 20–22 June 2011 in Brussels, and, in addition to regular sessions, will feature a public workshop on the afternoon of 22 June. It is to be noted that COST-TERRA meetings have an open participation policy; therefore, any researchers from around the world who work on developments of regulatory policies for (Continued on Newsletter page 4)
Global Communications Newsletter • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Report on the 12th International Conference on Communication Systems (ICCS 2010) By Michael Ong, Institute for Infocomm Research, Singapore The 12th International Conference on Communication Systems (ICCS 2010) was held at the Furama Riverfront Hotel, Singapore from 17–19 November 2010. This is a biennial series of conferences co-organized by the National University of Singapore (NUS) and the Institute for Infocomm Research (I2R), Singapore, and technically sponsored by IEEE Communications Society, IEEE Singapore Communications Chapter, and IEEE Singapore Vehicular Technology Chapter. ICCS 2010 was a three-day event comprising a keynote speech by a leading academic researcher, technical oral and poster presentations, and tutorials by industrial experts. This event continues to provide opportunities for researchers and practitioners to share their experience and ideas in the field of communications engineering and systems. The conference showcased a technical program consisting of 10 oral sessions, two poster sessions, five special sessions, and three tutorials covering many exciting aspects of wireless communications, optical communications, devices, and new emerging technologies. In particular, the five special sessions consisted of 25 invited papers, addressing the latest developments in the fields of broadband mobile communications, cognitive and cooperative communications, energy harvesting and sustainable communications, optical communica-
Clockwise from upper left: General Co-Chair Dr. Liang Yingchang at the opening ceremony, 18 November 2010; delegates touring the Singapore River on the way to the banquet dinner at the Asian Civilisation Museum; Dr. Sun Sumei, TPC Chair, introduced Prof. Fumiyuki Adachi for the keynote speech; local Arrangement Chair Dr Manjeet Singh toasting during the banquet dinner at the Asian Civilisation Museum.
(Continued on Newsletter page 4)
Highlights from the World Telecommunications Congress By Kohei Shiomoto, NTT, Japan The World Telecommunications Congress (WTC) builds on the traditions of quality, timeliness, and open interaction originating in the International Switching Symposium (ISS) and International Symposium on Subscriber Loops and Services (ISSLS). WTC brings together leading experts from industry, academia, and government to present the latest advances in technology and to exchange views on the future of telecommunications. After the merger of ISS and ISSLS, WTC has been held every two years: WTC 2006 and 2008 were held in Budapest, Hungary, and New Orleans, Louisiana (WTC 2008 was held in conjunction with IEEE GLOBECOM 2008). The last WTC was held in Vienna, Austria, on 13–14 September, 2010. In keynote sessions, six talks were presented by high-profile speakers from Europe, the United States, and Japan. The Congress was opened by Dr. Rüdiger Köster (TMobile Austria GmbH) with a presentation entitled “Telecommunications: The Infrastructure for the 21st Century.” Mag. Dr. August Reschreiter (Austrian Federal Ministry of Transport, Innovation and Technology) presented the talk “The Governmental Challenges in Light of Next Generation Networks.” And Ruprecht Niepold (European Commission, DG INFSO) presented his views on “The Digital Agenda for Europe — The
Opening session of WTC 2010. Policy and Regulatory Perspective.” On day two, Prof. Dale Hatfield (University of Colorado) presented “The Role of Self-Regulation in the Next Generation Network.” Dr. Kou Miyake (NTT, Japan) presented the talk “FTTH/NGN Service Deployment Strategy in NTT.” And Prof. Dr.-Ing. habil. Jörg Lange (Nokia Siemens Networks, Ger(Continued on Newsletter page 4)
Global Communications Newsletter • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
3 A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
ICCS 2010/continued from page 3 tions, and Underwater Communications. The conference also featured a keynote speech by IEEE Fellow Prof Fumiyuki Adachi from Tohoku University, Japan, titled “Challenges for Gigabit Wireless Pipe.” There were close to 300 submissions from more than 40 countries to the open call for papers, and 152 papers were accepted for presentation at the conference after a rigorous and challenging technical review process. For more information, please refer to iccs-2010.org/ index.htm
[email protected] ______ or Dr Michael Ong at ___________________
HIGHLIGHTS FROM WTC 2010/continued from page 3 many) gave a presentation entitled “Convergent Charging, Billing and Care – The Increasing Importance of Online Cost Control for Post-Paid Subscribers.” The regular technical sessions encompassed a wide variety of timely topics in telecommunications: Future Mobile, Network & Service Management, NGN, Mobile Ad Hoc, QoS, Ambient Assisted Living, Optical Networks, Mobile Access and PANs, Future Internet, Regulatory and Policy Issues, Service and Applications, and Security. Among quality papers, the best paper award and the best presentation award were selected. Dr. Florian Winkler (NEC Europe Ltd., Germany) received the best paper award for his paper entitled “Driving Mobile eCommerce Services Using Identity Management.” Dr. Erwin P. Rathgeb (Universität Duisburg-Essen, Germany) received the best presentation award for his lecture on “Security in the Net —Why Everything Used to Be Better, Bad Things Happen
A
BEMaGS F
Today and the Future Looks Bright.” WTC 2010 presented 54 talks in 12 sessions and was very successful, attracting participants from Germany, Austria, Japan, Italy, Poland, Hungary, Saudi Arabia, the United States, France, Spain, Turkey, Korea, Australia, the United Kingdom, Belgium, and Taiwan. The next WTC will be held in Miyazaki, Japan, on 5–6 March, 2012. WTC 2012 is sponsored by the IEICE and is technically co-sponsored by the VDE/ITG and the IEEE Communications Society. Please visit http://www.wtc2012.jp for details. See you in Japan!
LEBANON WORKSHOP/continued from page 2 Development Manager); Secunet/ISN-Technologies on Highly Secure Ethernet Encryption Concepts by Mr. Michael Frings (Regional Sales Director); Bank of America on Open Trust Frameworks for Online Security by Dr. Abbie Barbir (VP, Senior Security Architect); UN-ESCWA on Cyber Security Awareness by Dr. Nibal Idlebi (Chief of the ICT Applications Section); LU Faculty of Law on Regulatory Aspects of Cyber Security by Prof. Mona Al-Achkar; and Alfa on GSM Security Challenges by Mr. Issam El-Hajal (Head of Release and Portfolio Management Unit). The workshop activities included two coffee breaks and a lunch break which provided participants with the opportunity to network among each other and to follow up on the discussions with the specialist invited speakers. The IEEE Communications Society Lebanon Chapter Executive Committee has initiated planning activities for IEEE LCW ’11, which will take place in November 2011 with a focus on emergency communications. For more information about IEEE ComSoc Lebanon Chapter activities, check http://ewh.ieee.org/r8/lebanon/com
Global 4G MOBILE LICENSES/continued from page 1
Newsletter www.comsoc.org/pubs/gcn
STEFANO BREGNI Editor
MHz. Some virtual operators will create alliances to avoid losing this opportunity and thus address the investment power of the big operators.
Further Readings
Politecnico di Milano - Dept. of Electronics and Information Piazza Leonardo da Vinci 32, 20133 MILANO MI, Italy Ph.: +39-02-2399.3503 - Fax: +39-02-2399.3413 ___________
[email protected] _________ Email:
[email protected],
[1] http://www.mityc.es [2] M. Cave, C. Doyle, W. Webb, “Essentials of Modern Spectrum Management”, Cambridge Press, 2007 [3] http://ec.europa.eu/information_society/digital-agenda
IEEE COMMUNICATIONS SOCIETY
KHALED B. LETAIEF, VICE-PRESIDENT CONFERENCES SERGIO BENEDETTO, VICE-PRESIDENT MEMBER RELATIONS JOSÉ-DAVID CELY, DIRECTOR OF LA REGION GABE JAKOBSON, DIRECTOR OF NA REGION TARIQ DURRANI, DIRECTOR OF EAME REGION NAOAKI YAMANAKA, DIRECTOR OF AP REGION ROBERTO SARACCO, DIRECTOR OF SISTER AND RELATED SOCIETIES REGIONAL CORRESPONDENTS WHO CONTRIBUTED TO THIS ISSUE
ANA ALEJOS, SPAIN/USA (ANNALEJOS @UVIGO.ES) ____________ JOEL RODRIGUES, PORTUGAL (__________
[email protected]) __________________ KOHEI SHIOMOTO, JAPAN (SHIOMOTO
[email protected]) EWELL TAN, SINGAPORE (EWELL
[email protected]) ____________
®
A publication of the IEEE Communications Society
4
Communications IEEE
EUROPEAN RESEARCH/continued from page 2 CR/SDR technologies are welcome to come and present their research or simply listen in and join in the debates. Early research directions within COST-TERRA focused around the analysis and categorization of known CR/SDR use scenarios and business cases. Three parallel threads will be pursued for the time being: •CR/SDR deployment scenarios •CR/SDR coexistence studies •Economic aspects of CR/SDR regulation Later, the fourth research area will be activated to deal with the impact assessment of CR/SDR regulation. For more information on the aims, work programme, and ongoing results of COST-TERRA, please visit the action’s website at http://www.cost-terra.org/.
Global Communications Newsletter • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
EUROPE’S PREMIER MICROWAVE, RF, WIRELESS AND RADAR EVENT
A
BEMaGS F
_______________
European Microwave Week is the largest event dedicated to RF, Microwave, Radar and Wireless Technologies in Europe. Capitalising on the success of the previous shows, the event promises growth in the number of visitors and delegates. EuMW2011 will provide: • 7,500 sqm of gross exhibition space • 5,000 key visitors from around the globe • 1,700 - 2,000 conference delegates • In excess of 250 exhibitors Running alongside the exhibition are 3 separate, but complementary Conferences: • European Microwave Integrated Circuits Conference (EuMIC) • European Microwave Conference (EuMC) • European Radar Conference (EuRAD) Plus a one day Defence and Security Conference
Co-sponsored by:
European Microwave Association
Co-sponsored by:
Supported by:
Official Publication:
Organised by:
The 6th European Microwave Integrated Circuits Conference Co-sponsored by:
The 8th European Radar Conference
The 41st European Microwave Conference
Interested in exhibiting? Book online NOW!
www.eumweek.com For further information, please contact: Richard Vaughan Horizon House Publications Ltd. 16 Sussex Street London SW1V 4RW, UK E:
[email protected] _______________________ Tel: +44 20 7596 8742 Fax: +44 20 7596 8749
Communications IEEE
Kristen Anderson Horizon House Publications Inc. 685 Canton Street Norwood, MA 02062, USA E:
[email protected] _____________________ Tel: +1 781 769 9750 Fax: +1 781 769 5037
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
GUEST EDITORIAL
FUTURE INTERNET ARCHITECTURES: DESIGN AND DEPLOYMENT PERSPECTIVES
Raj Jain
T
Arjan Durresi
he last 40 years of research has matured packet switching technology as a key communication primitive. Its key use context, the Internet, has been phenomenally successful. From its humble beginning as a research network, it has evolved into a critical infrastructure for the development of businesses, societies, and nations. The Internet’s most popular application, the World Wide Web, has powered the present information age that has accelerated progress in all areas. There is no doubt that a lot has been achieved. Yet as we look toward the future, a very different set of research challenges present themselves. These challenges originate primarily from the “responsibilities” of handling an elite infrastructure, the “burden” of satisfying popular expectations, and catering to the “change” in its use context. The future Internet needs to cater to the responsibilities of a critical infrastructure. Security, energy efficiency, and performance guarantee are the primary issues. Also, the future Internet needs to live up to its “near-magical” perception of communication capabilities. It needs to be able to scale to billions of nodes and also provide support for the diversified requirements of next-generation applications. The original architecture of the Internet and its communication protocols were not designed for such requirements. Moreover, the use context for which the original Internet was designed has changed considerably. We have adapted to these changes through incremental modifications to the original architecture. On the one hand these changes have helped sustain the growth of the Internet while on the other it has increasingly made the Internet architecture brittle and non-deterministic. Thus, the basic underlying principles that have been instrumental in the Internet’s success need to be revisited and possibly redefined in light of future requirements. The networking research community has taken up the task for designing the architecture for the future Internet. Initially started as part of the FIND and GENI programs by the National Science Foundation (NSF) in the United States, future Internet research is now a key agenda for all leading research agencies around the world including the European Union, Japan, and China. The goal of this feature topic is to present some interesting
24
Communications IEEE
Subharthi Paul
design and deployment perspectives on the future Internet research. We received 48 papers. Six of these have been selected for publication in this issue. True to the spirit of diversified future Internet design, all six articles address different design and research areas. The topic should be interesting reading, with each article providing a new and fresh perspective on the design space. “A Survey of the Research on Future Internet Architectures” provides a concise and informative survey of the various next-generation Internet design initiatives around the world. The article is background reading for those who want a high-level look into the research landscape of diversified projects with very different objectives and design approaches. This article provides pointers for further research into specific projects and ideas to an interested reader. While we did not receive any papers from the four winners of the NSF Future Internet Architecture (FIA) competition, or any of the NSF GENI participants, this article provides brief insights into those projects. The reviews of this article were handled directly by Dr. Steve Gorshe, Editor-in-Chief of IEEE Communications Magazine. The article “Loci of Competition for Future Internet Architectures” proposes a new design principle for the future Internet that advocates “designing for competition.” This design principle is rooted in economics and represents a relevant interdisciplinary research area for future Internet design. The article identifies various “loci” in the design space that should allow for multiple competitive providers to coexist. This will prevent monopolies. Future design choices and innovations shall evolve more naturally based on market forces. The key challenges are to locate the proper loci across the horizontal and vertical design space that do not unnecessarily make the architecture too complex, and manage the interaction between the different loci to provide a seamless communication infrastructure. Clearly, the current Internet was not designed from the perspective of future commercial use. As a result, policy enforcements, security, and accountability across interorganizational boundaries have perennially been the problem areas for the current design. The “design for competition” principle will hopefully set the right economic circumstances for the design of the future Internet. Another interesting interdisciplinary research area that
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
GUEST EDITORIAL might potentially contribute new ideas to the design of a robust, self-managed, and naturally evolving future Internet design is biology. “Biological Principles for Future Internet Architecture Design” maps biological principles to network architectures. Biological systems present different models for intercommunications among elements to implement a self-sustaining system. These mechanisms may be modeled to design a robust and self-managed control plane for the future Internet. The article also provides a possible way to implement these abstract principles on actual networks. The key insight presented is that to correctly map a biological intercommunication mechanism to an equivalent networking mechanism, it is necessary to consider the scale and administrative control boundaries of the intercommunication scenario. One of the primary components of next-generation Internet research is a testbed platform. Designing testbeds for atscale experimentation is itself an independent area of research. The article “Enabling Future Internet Research: The FEDERICA Case” presents a discussion on the experiences with building and managing the FEDERICA testbed. Apart from a discussion on the testbed itself, its features, and the different technologies it uses, the article presents a list of projects that run on the testbed and how FEDERICA supports their diversified experimental contexts. While there is a lot of literature on testbed technologies, their properties, and their unique features, this article is differently organized such that the end user may be able to appreciate the different features of the testbed better through real use case examples. The next-generation Internet design space is highly diversified across different design philosophies, principles, and technologies. “Content, Connectivity, and Cloud: Ingredients for the Network of the Future” provides an integrated design framework from three of the most promising components of the nextgeneration Internet design space: information-centric network design, cloud computing, and open connectivity. The high point of this article is that it provides a lot of insight into each of these design space components and also integrates them into a coherent framework. A key area of research for next-generation network technologies is the programmable data plane. There are two aspects to this research. The first aspect is the set of control and management protocols that support the programmability of the underlying data plane. The second aspect is the system-level design of the high-performance programmable data plane itself. “PEARL: A Programmable Virtual Router Platform” presents a system-level design of a programmable data plane with discussion of both the hardware and software platforms. We would like to thank Editor-inChief Dr. Steve Gorshe and the IEEE publication staff for their assistance with this feature topic. Special thanks to 253 reviewers who took time to provide detailed comments. We are also thankful to the authors who submitted their articles for this special issue.
BIOGRAPHIES R AJ J AIN [F’93] (
[email protected]) ________ is a Fellow of ACM, and a winner of the ACM SIGCOMM Test of Time award, CDAC-ACCS Foundation Award 2009, and Hind Rattan 2011 award. He is currently a professor of computer science and engineering at Washington University in St. Louis. Previously, he was one of the co-founders of Nayna Networks, Inc., a next-generation telecommunications systems company in San Jose, California. He was a senior consulting engineer at Digital Equipment Corporation in Littleton, Massachusetts, and then a professor of computer and information sciences at Ohio State University , Columbus. He is the author of Art of Computer Systems Performance Analysis, which won the 1991 Best Advanced How-to Book, Systems award from the Computer Press Association. His fourth book, entitled High-Performance TCP/IP: Concepts, Issues, and Solutions, was published by Prentice Hall in November 2003. He recently co-edited Quality of Service Architectures for Wireless Networks: Performance Metrics and Management, published in April 2010. ARJAN DURRESI [S’03] (
[email protected]) _________ is currently an associate professor of computer and information science at Indiana University Purdue University Indianapolis (IUPUI). He was a senior software analyst at Telesoft Inc. in Rome, Italy; then a research scientist in computer and information sciences at Ohio State University; and an assistant professor of computer science at Louisiana State University, Baton Rouge. He has authored over 70 articles in journals and more than 140 in conferences in the fields of networking and security. He is the founder of several international workshops, including Heterogeneous Wireless Networks — HWISE in 2005, Advances in Information Security — WAIS in 2007, Bio and Intelligent Computing — BICOM 2008, and Trustworthy Computing — TWC in 2010. SUBHARTHI PAUL received his B.S. degree from the University of Delhi, India, and his Master’s degree in software engineering from Jadavpur University, Kolkata, India. He is presently a doctoral student in computer science and engineering at Washington University, St. Louis, Missouri. His primary research interests are in the area of future Internet architectures.
__________
__________
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
25
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
FUTURE INTERNET ARCHITECTURES: DESIGN AND DEPLOYMENT PERSPECTIVES
A Survey of the Research on Future Internet Architectures Jianli Pan, Subharthi Paul, and Raj Jain, Washington University
ABSTRACT The current Internet, which was designed over 40 years ago, is facing unprecedented challenges in many aspects, especially in the commercial context. The emerging demands for security, mobility, content distribution, etc. are hard to be met by incremental changes through ad-hoc patches. New clean-slate architecture designs based on new design principles are expected to address these challenges. In this survey article, we investigate the key research topics in the area of future Internet architecture. Many ongoing research projects from United States, the European Union, Japan, China, and other places are introduced and discussed. We aim to draw an overall picture of the current research progress on the future Internet architecture.
INTRODUCTION
This work was supported in part by a grant from Intel Corporation and NSF CISE Grant #1019119.
26
Communications IEEE
The Internet has evolved from an academic network to a broad commercial platform. It has become an integral and indispensable part of our daily life, economic operation, and society. However, many technical and non-technical challenges have emerged during this process, which call for potential new Internet architectures. Technically, the current Internet was designed over 40 years ago with certain design principles. Its continuing success has been hindered by more and more sophisticated network attacks due to the lack of security embedded in the original architecture. Also, IP’s narrow waist means that the core architecture is hard to modify, and new functions have to be implemented through myopic and clumsy ad hoc patches on top of the existing architecture. Moreover, it has become extremely difficult to support the ever increasing demands for security, performance reliability, social content distribution, mobility, and so on through such incremental changes. As a result, a clean-slate architecture design paradigm has been suggested by the research community to build the future Internet. From a non-technical aspect, commercial usage requires fine-grained security enforcement as opposed to the current “perimeter-based” enforcement.
0163-6804/11/$25.00 © 2011 IEEE
Security needs to be an inherent feature and integral part of the architecture. Also, there is a significant demand to transform the Internet from a simple “host-to-host” packet delivery paradigm into a more diverse paradigm built around the data, content, and users instead of the machines. All of the above challenges have led to the research on future Internet architectures. Future Internet architecture is not a single improvement on a specific topic or goal. A cleanslate solution on a specific topic may assume the other parts of the architecture to be fixed and unchanged. Thus, assembling different cleanslate solutions targeting different aspects will not necessarily lead to a new Internet architecture. Instead, it has to be an overall redesign of the whole architecture, taking all the issues (security, mobility, performance reliability, etc.) into consideration. It also needs to be evolvable and flexible to accommodate future changes. Most previous clean-slate projects were focused on individual topics. Through a collaborative and comprehensive approach, the lessons learned and research results obtained from these individual efforts can be used to build a holistic Internet architecture. Another important aspect of future Internet architecture research is the experimentation testbeds for new architectures. The current Internet is owned and controlled by multiple stakeholders who may not be willing to expose their networks to the risk of experimentation. So the other goal of future Internet architecture research is to explore open virtual large-scale testbeds without affecting existing services. New architectures can be tested, validated, and improved by running on such testbeds before they are deployed in the real world. In summary, there are three consecutive steps leading toward a working future Internet architecture: Step 1: Innovations in various aspects of the Internet Step 2: Collaborative projects putting multiple innovations into an overall networking architecture Step 3: Testbeds for real-scale experimentation It may take a few rounds or spirals to work out a
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
future Internet architecture that can fit all the requirements. Future Internet research efforts may be classified based on their technical and geographical diversity. While some of the projects target at individual topics, others aim at holistic architectures by creating collaboration and synergy among individual projects. Research programs specifically aimed at the design of the future Internet have been set up in different countries around the globe, including the United States, the European Union (EU), Japan, and China. The geographical diversity of research presents different approaches and structures of these different research programs. While dividing the projects by their major topics is also possible, due to the holistic architecture goals, different projects may have some overlap. Over the past few years’ future Internet research has gathered enormous momentum as evidenced by the large number of research projects in this area. In this article, primarily based on the geographical diversity, we present a short survey limited in scope to a subset of representative projects and discuss their approaches, major features, and potential impact on the future. We discuss the key research topics and design goals for the future Internet architectures. Research projects in the United States, European Union, and Asian countries are discussed in detail, respectively. Some of our discussions and perspectives on future Internet architectures are included later. Finally, a summary concludes the article.
KEY RESEARCH TOPICS In this section, we discuss some key research topics that are being addressed by different research projects. Content- or data-oriented paradigms: Today’s Internet builds around the “narrow waist” of IP, which brings the elegance of diverse design above and below IP, but also makes it hard to change the IP layer to adapt for future requirements. Since the primary usage of today’s Internet has changed from host-to-host communication to content distribution, it is desirable to change the architecture’s narrow waist from IP to the data or content distribution. Several research projects are based on this idea. This category of new paradigms introduces challenges in data and content security and privacy, scalability of naming and aggregation, compatibility and co-working with IP, and efficiency of the new paradigm. Mobility and ubiquitous access to networks: The Internet is experiencing a significant shift from PC-based computing to mobile computing. Mobility has become the key driver for the future Internet. Convergence demands are increasing among heterogeneous networks such as cellular, IP, and wireless ad hoc or sensor networks that have different technical standards and business models. Putting mobility as the norm instead of an exception of the architecture potentially nurtures future Internet architecture with innovative scenarios and applications. Many collaborative research projects in academia and industry are pursuing such research topics with great interest.
These projects also face challenges such as how to trade off mobility with scalability, security, and privacy protection of mobile users, mobile endpoint resource usage optimization, and so on. Cloud-computing-centric architectures: Migrating storage and computation into the “cloud” and creating a “computing utility” is a trend that demands new Internet services and applications. It creates new ways to provide global-scale resource provisioning in a “utilitylike” manner. Data centers are the key components of such new architectures. It is important to create secure, trustworthy, extensible, and robust architecture to interconnect data, control, and management planes of data centers. The cloud computing perspective has attracted considerable research effort and industry projects toward these goals. A major technical challenge is how to guarantee the trustworthiness of users while maintaining persistent service availability. Security: Security was added into the original Internet as an additional overlay instead of an inherent part of the Internet architecture. Now security has become an important design goal for the future Internet architecture. The research is related to both the technical context and the economic and public policy context. From the technical aspect, it has to provide multiple granularities (encryption, authentication, authorization, etc.) for any potential use case. Also, it needs to be open and extensible to future new security related solutions. From the non-technical aspect, it should ensure a trustworthy interface among the participants (e.g., users, infrastructure providers, and content providers). There are many research projects and working groups related to security. The challenges on this topic are very diverse, and multiple participants make the issue complicated. Experimental testbeds: As mentioned earlier, developing new Internet architectures requires large-scale testbeds. Currently, testbed research includes multiple testbeds with different virtualization technologies, and the federation and coordination among these testbeds. Research organizations from the United States, European Union, and Asia have initiated several programs related to the research and implementation of large-scale testbeds. These projects explore challenges related to large-scale hardware, software, distributed system test and maintenance, security and robustness, coordination, openness, and extensibility. Besides these typical research topics, there are several others, including but not limited to networked multimedia; “smart dust,” also called the “Internet of things”; and Internet services architecture. However, note that in this survey, we are not trying to enumerate all the possible topics and corresponding research projects. Instead, we focus on a representative subset and discuss a few important ongoing research projects. Due to length limitations, we are not able to enumerate all the references for the projects discussed below. However, we do have a longer survey [18], which includes a more complete reference list for further reading.
IEEE
BEMaGS F
Since the primary usage of the today’s Internet has changed from host-to-host communication to content distribution, it is desirable to change the architecture’s narrow waist from IP to the data or content distribution.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
27
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Categories
Project or cluster names (selected)
FIA
NDN, MobilityFirst, NEBULA, XIA, etc.
FIND
CABO, DAMS, Maestro, NetSerV, RNA, SISS, etc. (more than 47 total) Spiral1: (5 clusters totally): DETER (1 project), PlanetLab (7 projects), ProtoGENI (5 projects), ORCA (4 projects), ORBIT (2 projects; 8 not classified; 2 analysis projects
GENI Spiral2: over 60 active projects as of 2009* Spiral3: about 100 active projects as of 2011* * GENI design and prototyping projects can last for more than one spiral.
Table 1. U.S. projects and clusters on the future Internet.
RESEARCH PROJECTS FROM THE UNITED STATES Research programs on future Internet architecture in United States are administrated by the National Science Foundation (NSF) directorate for Computer and Information Science and Engineering (CISE).
FIA AND FIND The Future Internet Architecture (FIA) program [1] of the National Science Foundation (NSF) is built on the previous program, Future Internet Design (FIND) [2]. FIND funded about 50 research projects on all kinds of design aspects of the future Internet. FIA is the next phase to pull together the ideas into groups of overall architecture proposals. There are four such collaborative architecture groups funded under this program, and we introduce them here. Table 1 illustrates the overall research projects from the United States, including FIA and FIND. Named Data Networking (NDN) — The Named Data Networking (NDN) [3] project is led by the University of California, Los Angeles with participation from about 10 universities and research institutes in the United States. The initial idea of the project can be traced to the concept of content-centric networks (CCNs) by Ted Nelson in the 1970s. After that, several projects such as TRIAD at Stanford and DONA from the University of California at Berkeley were carried out exploring the topic. In 2009 Xerox Palo Alto Research Center (PARC) released the CCNx project led by Van Jacobson, who is also one of the technical leaders of the NDN project. The basic argument of the NDN project is that the primary usage of the current Internet has changed from end-to-end packet delivery to a content-centric model. The current Internet, which is a “client-server” model, is facing challenges in supporting secure content-oriented functionality. In this information dissemination model, the network is “transparent” and just forwarding data (i.e., it is “content-unaware”). Due to this unawareness, multiple copies of the same data are sent between endpoints on the
28
Communications IEEE
A
BEMaGS F
network again and again without any traffic optimization on the network’s part. The NDN uses a different model that enables the network to focus on “what” (contents) rather than “where” (addresses). The data are named instead of their location (IP addresses). Data become the first-class entities in NDN. Instead of trying to secure the transmission channel or data path through encryption, NDN tries to secure the content by naming the data through a security-enhanced method. This approach allows separating trust in data from trust between hosts and servers, which can potentially enable content caching on the network side to optimize traffic. Figure 1 is a simple illustration of the goal of NDN to build a “narrow waist” around content chunks instead of IP. NDN has several key research issues. The first one is how to find the data, or how the data are named and organized to ensure fast data lookup and delivery. The proposed idea is to name the content by a hierarchical “name tree” which is scalable and easy to retrieve. The second research issue is data security and trustworthiness. NDN proposes to secure the data directly instead of securing the data “containers” such as files, hosts, and network connections. The contents are signed by public keys. The third issue is the scaling of NDN. NDN names are longer than IP addresses, but the hierarchical structure helps the efficiency of lookup and global accessibility of the data. Regarding these issues, NDN tries to address them along the way to resolve the challenges in routing scalability, security and trust models, fast data forwarding and delivery, content protection and privacy, and an underlying theory supporting the design. MobilityFirst — The MobilityFirst [4] project is led by Rutgers University with seven other universities. The basic motivation of MobilityFirst is that the current Internet is designed for interconnecting fixed endpoints. It fails to address the trend of dramatically increasing demands of mobile devices and services. The Internet usage and demand change is also a key driver for providing mobility from the architectural level for the future Internet. For the near term, MobilityFirst aims to address the cellular convergence trend motivated by the huge mobile population of 4 to 5 billion cellular devices; it also provides mobile peer-to-peer (P2P) and infostation (delay-tolerant network [DTN]) application services which offer robustness in case of link/network disconnection. For the long term, in the future, MobilityFirst has the ambition of connecting millions of cars via vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) modes, which involve capabilities such as location services, georouting, and reliable multicast. Ultimately, it will introduce a pervasive system to interface human beings with the physical world, and build a future Internet around people. The challenges addressed by MobilityFirst include stronger security and trust requirements due to open wireless access, dynamic association, privacy concerns, and greater chance of network failure. MobilityFirst targets a clean-slate design
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
directly addressing mobility such that the fixed Internet will be a special case of the general design. MobilityFirst builds the “narrow waist” of the protocol stack around several protocols: • Global name resolution and routing service • Storage-aware (DTN-like) routing protocol • Hop-by-hop segmented transport • Service and management application programming interfaces (APIs) The DTN-like routing protocol is integrated with the use of self-certifying public key addresses for inherent trustworthiness. Functionalities such as context- and location-aware services fit into the architecture naturally. An overview of the MobilityFirst architecture is shown in Fig. 2. It shows all the building blocks mentioned above and how they work together. Some typical research challenges of MobilityFirst include: • Trade-off between mobility and scalability • Content caching and opportunistic data delivery • Higher security and privacy requirements • Robustness and fault tolerance NEBULA — NEBULA [5] is another FIA project focused on building a cloud-computing-centric network architecture. It is led by the University of Pennsylvania with 11 other universities. NEBULA envisions the future Internet consisting of a highly available and extensible core network interconnecting data centers to provide utility-like services. Multiple cloud providers can use replication by themselves. Clouds comply with the agreement for mobile “roaming” users to connect to the nearest data center with a variety of access mechanisms such as wired and wireless links. NEBULA aims to design the cloud service embedded with security and trustworthiness, high service availability and reliability, integration of data centers and routers, evolvability, and economic and regulatory viability. NEBULA design principles include: • Reliable and high-speed core interconnecting data centers • Parallel paths between data centers and core routers • Secure in both access and transit • A policy-based path selection mechanism • Authentication enforced during connection establishment With these design principles in mind, the NEBULA future Internet architecture consists of the following key parts: • The NEBULA data plane (NDP), which establishes policy-compliant paths with flexible access control and defense mechanisms against availability attacks • NEBULA virtual and extensible networking techniques (NVENT), which is a control plane providing access to applicationselectable service and network abstractions such as redundancy, consistency, and policy routing • The NEBULA core (NCore), which redundantly interconnects data centers with ultrahigh-availability routers NVENT offers control plane security with policy-selectable network abstraction including mul-
IEEE
BEMaGS F
A “narrow waist” around content chunks instead of the IP. Web, email, VoIP, eBusiness...
Browsers, Skype, online gaming...
HTTP, RTP, SMTP...
File streams...
TCP, UDP, SCTP...
Security...
IP
“Narrow waist”
Contents
Ethernet, WIFi...
Strategies...
CSMA, ADSL, Sonet...
P2P, UDP, IP broadcast...
Optical fiber, copper, radio...
Optical fiber, copper, radio...
Figure 1. The new “narrow waist” of NDN (right) compared to the current Internet (left). Mobility first routers are with storage capability Hop-by-hop segment transport
Core network
Generalized DTN routing
Data plane Global name resolution service
Name to address mapping
Control and management plane
Figure 2. MobilityFirst architecture. tipath routing and use of new networks. NDP involves a novel approach for network path establishment and policy-controlled trustworthy paths establishment among NEBULA routers. Figure 3 shows the NEBULA architecture comprising the NDP, NVENT, and NCore, and shows how they interact with each other. eXpressive Internet Architecture (XIA) — Expressive Internet Architecture (XIA) [6] is also one of the four projects from the NSF FIA program, and was initiated by Carnegie Mellon University collaborating with two other universities. As we observe, most of the research projects on future Internet architectures realize the importance of security and consider their architecture carefully to avoid the flaws of the original Internet design. However, XIA directly and explicitly targets the security issue within its design. There are three key ideas in the XIA architecture: • Define a rich set of building blocks or communication entities as network principals including hosts, services, contents, and future additional entities. • It is embedded with intrinsic security by using self-certifying identifiers for all principals for integrity and accountability properties.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
29
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
NDP path Wireless access network
NVENT
NDP path NVENT Reliable Wired trustworthy access core network Data center network (NCore)
Data center Transit network
Figure 3. NEBULA architecture components and their interactions. • A pervasive “narrow waist” (not limited to the host-based communication as in the current Internet) for all key functions, including access to principals, interaction among stakeholders, and trust management; it aims to provide interoperability at all levels in the system, not just packet forwarding. The XIA components and their interactions are illustrated in Fig. 4. The core of the XIA is the Expressive Internet Protocol (XIP) supporting communication between various types of principals. Three typical XIA principal types arecontent, host (defined by “who”), and service (defined by what it does). They are open to future extension. Each type of principal has a narrow waist that defines the minimal functionality required for interoperability. Principles talk to each other using expressive identifiers (XIDs), which are 160 bit identifiers identifying hosts, pieces of content, or services. The XIDs are basically self-certifying identifiers taking advantage of cryptographic hash technology. By using this XID, the content retrieval no longer relies on a particular host, service or network path. XIP can then support future functions as a diverse set of services. For low-level services, it uses a path-segment-based network architecture (named Tapa in their previous work) as the basic building block; and builds services for content-transfer and caching and service for secure content provenance at a higher level. XIA also needs various trustworthy mechanisms and provides network availability even when under attack. Finally, XIA defines explicit interfaces between network actors with different roles and goals.
GLOBAL ENVIRONMENT FOR NETWORK INNOVATIONS (GENI) GENI [7] is a collaborative program supported by NSF aimed at providing a global large-scale experimental testbed for future Internet architecture test and validation. Started in 2005, it has attracted broad interest and participation from both academia and industry. Besides its initial support from existing projects on a dedicated backbone network infrastructure, it also aims to attract other infrastructure platforms to participate in the federation — the device control framework to provide these participating networks with users and operating environments, to observe, measure, and record the resulting experimental outcomes. So generally, GENI is differ-
30
Communications IEEE
A
BEMaGS F
ent from common testbeds in that it is a generalpurpose large-scale facility that puts no limits on the network architectures, services, and applications to be evaluated; it aims to allow clean-slate designs to experiment with real users under real conditions. The key idea of GENI is to build multiple virtualized slices out of the substrate for resource sharing and experiments. It contains two key pieces: • Physical network substrates that are expandable building block components • A global control and management framework that assembles the building blocks together into a coherent facility Thus, intuitively two kinds of activities will be involved in GENI testbeds: one is deploying a prototype testbed federating different small and medium ones together (e.g., the OpenFlow testbed for campus networks [8]); the other is to run observable, controllable, and recordable experiments on it. There are several working groups concentrating on different areas, such as the control framework working group; GENI experiment workflow and service working group; campus/operation, management, integration, and security working group; and instrumentation and management working group. The GENI generic control framework consists of several subsystems and corresponding basic entities: • Aggregate and components • Clearinghouse • Research organizations, including researchers and experiment tools • Experiment support service • “Opt-in” end users • GENI operation and management Clearinghouses from different organizations and places (e.g., those from the United States and European Union) can be connected through federation. By doing this, GENI not only federates with identical “GENI-like” systems, but also with any other system if they comply with a clearly defined and relatively narrow set of interfaces for federation. With these entities and subsystems, “slices” can be created on top of the shared substrate for miscellaneous researchdefined specific experiments, and end users can “opt in” onto the GENI testbed accordingly. GENI’s research and implementation plan consists of multiple continuous spirals (currently in spiral 3). Each spiral lasts for 12 months. Spiral 1 ran from 2008 to 2009; spiral 2 ran from 2009 to 2010; spiral 3 started in 2011. In spiral 1, the primary goals were to demonstrate one or more early prototypes of the GENI control framework and end-to-end slice operation across multiple technologies; there were five competing approaches to the GENI control framework, called “clusters.” Cluster A was the Trial Integration Environment based on DETER (TIED) control framework focusing on federation, trust, and security. It was a one-project cluster based on the CyberDefense Technology Experimental Research (DETER) control framework by the University of Southern California (USC)/ISI, which is an individual “mini-GENI” testbed to demonstrate
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
F
Users
Application Services
Host support
Content support
Services support
eXpressive Internet protocol
Intrinsic security
Figure 4. XIA components and interactions.
We can see that spirals 1 and 2 integrated a very wide variety of testbeds into its control framework. Spiral 2 was the second phase aiming to move toward continuous experimentation. Key developments include improved integration of GENI prototypes; architecture, tools, and services enabling experiment instrumentation; interoperability across GENI prototypes; and researcher identity management. In spiral 3, the goal is to coordinate the design and deployment of a first GENI Instrumentation and Measurement Architecture. Supporting experimental use of GENI and making it easier to use are also key goals. Also, more backbone services and participants are expected to join in the GENI framework for this spiral. Another notable and unique characteristic offered by GENI is that instrumentation and measurement support have been designed into the system from the beginning since the ultimate goal of GENI is to provide an open and extensible testbed for experimentation with various new Internet architectures.
RESEARCH PROJECTS FROM THE EUROPEAN UNION AND ASIA The European Union has also initiated a bundle of research projects on future Internet architectures. In this section, we introduce the research organized under the European Seventh Framework Programme (FP7) along with that in Japan and China.
EUROPEAN UNION The European Future Internet Assembly [19] (abbreviated FIA as in the United States) is a collaboration between projects under FP7 on future Internet research. Currently, the FIA brings together about 150 projects that are part of FP7. These projects have a wide coverage, including the network of the future, cloud computing, Internet of service, trustworthy information and communication technology (ICT), networked media and search systems, socio-economic aspects of the future Internet, application domain, and Future Internet Research and Experimentation (FIRE) [10]. The FIA maintains a European Future Internet Portal [20],
IEEE Communications Magazine • July 2011
IEEE
BEMaGS
Trustworthy network operation
federated and coordinated network provisioning. Cluster A particularly aimed to provide usability across multiple communities through federation. The project delivered software “fedd” as the implementation of the TIED federation architecture providing dynamic and on-demand federation, and interoperability across ProtoGENI, GENIAPI, and non-GENI aggregate. It included an Attribute Based Access Control (ABAC) mechanism for large-scale distributed systems. It created a federation with two other projects: StarBED in Japan and ProtoGENI in the United States. Cluster B was a control framework based on PlanetLab implemented by Princeton University emphasizing experiments with virtualized machines over the Internet. By the end of spiral 2, it included at least 12 projects from different universities and research institutes. The results of these projects are to be integrated into the PlanetLab testbed. PlanetLab provided “GENIwrapper” code for independent development of an aggregate manager (AM) for Internet entities. A special “lightweight” protocol was introduced to interface PlanetLab and OpenFlow equipment. Through these mechanisms, other projects in the cluster can design their own substrates and component managers with different capacities and features. Cluster C was the ProtoGENI control framework by the University of Utah based on Emulab, emphasizing network control and management. By the end of spiral 2, it consisted of at least 20 projects. The cluster integrated these existing and under-construction systems to provide key GENI functions. The integration included four key components: a backbone based on Internet2; sliceable and programmable PCs and NetFPGA cards; and subnets of wireless and wired edge clusters. Cluster C so far is the largest set of integrated projects in GENI. Cluster D was Open Resource Control Architecture (ORCA) from Duke University and RENCI focusing on resource allocation and integration of sensor networks. By the end of spiral 2, it consisted of five projects. ORCA tried to include optical resources from the existing Metro-Scale Optical Testbed (BEN). Different from other clusters, the ORCA implementation included the integration of wireless/sensor prototypes. It maintains a clearinghouse for the testbeds under the ORCA control framework through which it connects to the national backbone and is available to external researchers. Cluster E was Open-Access Research Testbed for Next-Generation Wireless Networks (ORBIT) by Rutgers University focusing on mobile and wireless testbed networks. It included three projects by the end of spiral 2. The basic ORBIT did not include a full clearinghouse implementation. Cluster E tried to research how mobile and wireless work can affect and possibly be merged into the GENI architecture. WiMAX is one of the wireless network prototypes in this cluster. A more detailed description of the clusters and their specific approaches and corresponding features can be found in our previous survey [18]. Even more details can be found from GENI project websites and wikis [7].
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
User-network
IEEE
Network-network
Communications
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
31
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A significant trait of the “Network of the
Categories
Project names (selected)
Future Architectures and Technologies
4AWARD, TRILOGY, EIFFEL, SPARC, SENSEI, Socrates, CHANGE, PSIRP, etc.
Services, Software, and Virtualization
ALERT, FAST, PLAY, S-Cube, SLA@SOI, VISION Cloud, etc.
Network Media
3DLife, COAST, COMET, FutureNEM, nextMEDIA, P2P-Next, etc.
Internet of Things
ASPIRE, COIN, CuteLoop, SYNERGY, etc.
Trustworthiness
ABC4Trust, AVANTSSAR, ECRYPT II, MASTER, uTRUSTit, etc.
Testbeds
FIRE, N4C, OPNEX, OneLAB2, PII, WISEBED, G-Lab, etc.
Others
HYDRA, INSPIRE, SOCIALNETS, etc.
Future” is that the research projects
A
BEMaGS F
cover a very wide range of topics and a number of commercial organizations, including traditional telecommunication companies, participate in
Table 2. EU research projects on future Internet.
the research consortiums.
which is an important web portal for sharing information and interaction among the participating projects. Multiple FIA working groups have been formed to encourage collaboration among projects. Of these projects, around 90 of them were launched following the calls of FP7 under the “Network of the Future” Objective 1.1. They can be divided into three clusters: “Future Internet Technologies (FI),” “Converged and Optical Networks (CaON),” and “Radio Access and Spectrum (RAS).” The total research funding since 2008 is over €390 million. A subset of the projects is shown in Table 2. A significant trait of the “Network of the Future” [17] is that the research projects cover a very wide range of topics and a number of commercial organizations, including traditional telecommunication companies, participate in the research consortiums. Since there are a large number of projects, we selected a few representative ones and explain them in some detail. They are all under FP7 based on a series of design objectives categorized by the ICT challenge #1 of building “Pervasive and Trusted Network and Service Infrastructure.” Due to the large number of projects, for the architecture research in this article, we selected a project named 4WARD (Architecture and Design for the Future Internet), and for the testbed we selected FIRE. We selected them due to the fact that FIRE is often deemed the European counterpart project to GENI, and the 4WARD project aims at a general architectural level of redesign of the Internet, and we feel that it is representative of the rest. It also involves a large number of institutions’ participation and cooperation. In the following, we discuss these two projects briefly. 4WARD — 4WARD [9] is an EU FP7 project on designing a future Internet architecture led primarily by an industry consortium. The funding is over 45 million dollars for a 2-year period. The key 4WARD design goals are: • To create a new “network of information” paradigm in which information objects have their own identity and do not need to be
32
Communications IEEE
bound to hosts (somewhat similar to the goal of the NDN project) • To design the network path to be an active unit that can control itself and provide resilience and failover, mobility, and secure data transmission • To devise “default-on” management capability that is an intrinsic part of the network itself • To provide dependable instantiation and interoperation of different networks on a single infrastructure. Thus, on one hand, 4WARD promotes the innovations needed to improve a single network architecture; on the other hand, it enables multiple specialized network architectures to work together in an overall framework. There are five task components in the 4WARD research: • A general architecture and framework • Dynamic mechanisms for securely sharing resources in virtual networks • “Default-on” network management system; a communication path architecture with multipath and mobility support • Architecture for information-oriented networks Note that 4WARD is one of many projects under the FP7 framework on future Internet architecture research. Readers can find more information on a complete list of the projects from [19]. Some typical projects focusing on different aspects of future architecture are listed in Table 2. Future Internet Research and Experimentation (FIRE) — FIRE [10] is one of the European Union’s research projects on testbeds and is like a counterpart of GENI in the United States. FIRE was started in 2006 in FP6 and has continued through several consecutive cycles of funding. FIRE involves efforts from both industry and academia. It is currently in its “third wave” focusing on providing federation and sustainability between 2011 and 2012. Note that the FIRE project’s research is built on the previous work on the GEANT2 (Gigabit European Academic Networking Technology) project [11], which is the infrastructure testbed connecting over 3000 research organizations in Europe.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
FIRE has two interrelated dimensions: • To support long-term experimentally driven research on new paradigms and concepts and architectures for the future Internet • To build a large-scale experimentation facility by gradually federating existing and future emerging testbeds FIRE also expects not only to change the Internet in technical aspects but also in socio-economic terms by treating socio-economic requirements in parallel with technical requirements. A major goal of FIRE is federation, which by definition is to unify different self-governing testbeds by a central control entity under a common set of objectives. With this goal in mind, the FIRE project can be clustered in a layered way as depicted in Fig. 5. As shown in the figure, it contains three basic clusters. The top-level cluster consists of a bundle of novel individual architectures for routing and transferring data. The bottom cluster consists of projects providing support for federation. In the middle is the federation cluster, which consists of the existing testbeds to be federated. These existing small and medium-sized testbeds can be federated gradually to meet the requirements for emerging future Internet technologies. Documents describing these sub-testbeds can be found on the FIRE project website [10].
ASIA Asian countries such as Japan and China also have projects on future Internet architectures. Japan — Japan has broad collaborations with both the United States and European Union regarding future Internet research. It participates in PlanetLab in the United States, and the testbed in Japan is also federated with the German G-Lab facility. The Japanese research program on future Internet architecture is called New Generation Network (NWGN) sponsored by the Japan National Institute of Information and Communications Technology (NICT). The Japanese research community defines the cleanslate architecture design as “new generation” and the general IP-based converged design as “next generation” (NXGN) design. NWGN started in June 2010 and expects to change the network technologies and Internet community with broad impact in both the short term (to 2015) and long term (to 2050). Like the projects in the United States and European Union, NWGN consists of a series of sub-projects collaborated on by academia and industry. The sub-projects range from architecture designs, testbed designs, virtualization laboratories, and wireless testbeds to data-centric networking, service-oriented networks, advanced mobility management over network virtualization, and green computing. Rather than enumerating all projects, we briefly discuss the architecture project AKARI [12] and the testbed projects JGN2plus [13] and JGN-X (JGN stands for Japan Gigabit Network). The reason we selected these projects is similar to the reason we selected FIA and GENI. AKARI is so far the biggest architectural research project in Japan; JGN2plus and JGN-X are the testbed research counterparts to GENI and FIRE.
IEEE
BEMaGS F
Experimentally-driven, multi-disciplinary research
Testbeds
VITAL++
WISEBED
PII Federica OneLab2
Support actions
Figure 5. FIRE clustering of projects.
AKARI: AKARI means “a small light in the darkness.” The goal of AKARI is a clean-slate approach to design a network architecture of the future based on three key design principles: • “Crystal synthesis,” which means to keep the architecture design simple even when integrating different functions • “Reality connected,” which separates the physical and logical structures • “Sustainable and evolutional,” which means it should embed the “self-*” properties (self-organizing, self-distributed, self-emergent, etc.), and be flexible and open to the future changes AKARI is supposed to assemble five subarchitecture models to become a blueprint NWGN: • An integrated subarchitecture based on a layered model with cross-layer collaboration; logical identity separate from the data plane (a kind of ID/locator split structure) • A subarchitecture that simplifies the layered model by reducing duplicated functions in lower layers • A subarchitecture for quality of service (QoS) guarantee and multicast • A subarchitecture to connect heterogeneous networks through virtualization • A mobile access subarchitecture for sensor information distribution and regional adaptive services AKARI is currently in the process of a proofof-concept design and expects to get a blueprint in 2011. Through systematic testbed construction and experimentations, it aims to establish a new architecture ready for public deployment by 2016. JGN2plus and JGN-X: JGN2plus is the nationwide testbed for applications and networks in Japan, and also the testbed for international federation. It includes broad collaboration from both industry and academia. It evolved as JGN, migrated to JGN II in 2004, and then to JGN2plus in 2008. From 2011, the testbed is under JGN-X, which targets to be the real NWGN testbed to deploy and validate AKARI research results. JGN2plus provides four kinds of services:
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
33
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Any architecture that requires investment without immediate payoff is bound to fail. Of course, the payoff will increase as the deployment of the new technology increases, economies of scale reduce the cost, and eventually the old architecture deployed base will diminish and disappear.
34
Communications IEEE
• Layer 3 (L3) IP connection • L2 Ethernet connection • Optical testbed • Overlay service platform provisioning There are also five subtopics in the research of JGN2plus: • NWGN service platform fundamental technologies • NWGN service testbed federation technology • Middleware and application of lightpath NWGN • Component establishment technologies for NWGN operation • Verification of new technologies for international operation JGN2plus expects to create collaboration among industry, academia, and government for NWGN experiments. It also aims to contribute to human resource development in the ICT area via these experiments. China — The research projects on future Internet in China are mostly under the 863 Program, 973 Program, and “12th Five-Year Plan Projects” administrated by the Ministry of Science and Technology (MOST) of China. Currently there are several ongoing research projects, which include: • New Generation Trustworthy Networks (from 2007 to 2010) • New Generation Network Architectures (from 2009 to 2013) • Future Internet Architectures (from 2011 to 2015) Project 1 was still IP - network research instead of a clean-slate future Internet. It consists of research sub-projects on new network architecture, next generation broadcasting (NGB), new network services, a national testbed for new generation networks and services, new routing/ switching technology, a new optical transmission network, and low-cost hybrid access equipment. Besides the research projects on future Internet architecture, there are also ongoing research projects for building a China Next Generation Internet (CNGI) testbed. It is based on the previous infrastructure network testbed of the China Education and Research Network (CERNET [14] and CERNET2 [15]) and the China Science and Technology Network (CSTNET). A terabit optical, terabit WDM, terabit router plus IPTV testbed called (3T-NET) was also announced on July 2009 as NGB. The testbed projects are mostly industry oriented with specific interest in IPv6 related protocols and applications. We observed that the current future Internet architecture research in China leans heavily toward IPv6 related testbed, which is relatively short-term. To some extent, it reveals the pain China felt due to the collision between the extreme shortage of IPv4 address space in China and the ever expanding demands from increasing customers and novel services. Longer-term research projects on innovative architectural research are still in the cradle compared to those of the United States and European Union.
A
BEMaGS F
DISCUSSIONS AND PERSPECTIVES Having presented a variety of research projects, we find that there are several issues worth discussing. In this section, we give our perspective regarding these issues. Of course, there is no agreement among researchers regarding these perspectives, and none is implied. Clean-slate vs. evolutionary: Clean-slate designs impose no restriction and assumption on the architectural design. The key idea is not to be subjected to the limitations of the existing Internet architecture. It is also called “new generation” by Japanese and Chinese researchers. While the architectures can be revolutionary, their implementation has to be evolutionary. Today, the Internet connects billions of nodes and has millions of applications that have been developed over the last 40 years. We believe any new architecture should be designed with this reality in mind; otherwise, it is bound to fail. Legacy nodes and applications should be able to communicate over the new architecture without change (with adapter nodes at the boundary), and new nodes and applications should similarly be able to communicate over the existing Internet architecture. Of course, the services available to such users will be an intersection of those offered by both architectures. Also, the new architecture may provide adaptation facilities for legacy devices at their boundary points. Various versions of Ethernet are good examples of such backward compatibility. Some variations of IP are potential examples of missing this principle. New architecture deployment will start in a very small scale compared to the current Internet. These early adopters should have economic incentives for change. Any architecture that requires investment without immediate payoff is bound to fail. Of course, the payoff will increase as the deployment of the new technology increases, economies of scale reduce the cost and eventually the old architecture deployed base will diminish and disappear. Integration of security, mobility, and other functionalities: It is well understood and agreed that security, mobility, self-organization, disruption tolerance, and so on are some of the key required features for the future Internet. However, most of the projects, even for those collaborative ones like in FIA program, put more emphasis on a specific attribute or a specific set of problems. It seems to be a tough problem to handle many challenges in a single architecture design. Currently, for the collaborative projects such as FIA, they are trying to integrate miscellaneous previous research results into a coherent one trying to balance some of the issues. Although different projects have different emphases, it is beneficial to create such diversity and allow a bunch of integrated architectures to potentially compete in the future. However, we believe that there is still a long way to go before there is a next-generation architecture unifying these different lines of designs. For example, we observe that the four U.S. FIA projects concentrate on four different specific issues. Self-certifying and hash-based addresses are effective tools for security. However, securi-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
ty needs much more consideration on both micro and macro scopes. Content- and information-centric features are also important trends, but how to integrate these differing requirements and resulting architectures is still a pending problem. We expect that more integration research will be required when such issues emerge in the future. It is therefore desirable for different projects to create some synergy for the integration process. Architectures built around people instead of machines: It has been widely realized that the usage pattern of the Internet has changed, and the trend of building future Internet architecture around the contents, data, and users seems to be justifiable and promising in the future. Design goal changes naturally lead to design principle changes. Different patterns may emerge without any further synthesis. Current existing projects on future Internet architectures sort out different principles according to their own design emphases. From our perspective, it is essential and important to form a systematic and comprehensive theory in the research process rather than designing based only on experiences. It may take several continuous spirals between theoretical improvement and practical experience to achieve a sound architecture. We believe more research in this area may be desirable and meaningful for future Internet research. Interfaces among stakeholders: Future Internet architectures are required to provide extensible and flexible explicit interfaces among multiple stakeholders (users, Internet service providers, application service providers, data owners, and governments) to allow interaction, and enforce policies and even laws. A typical example is Facebook, which creates a complex situation for data, privacy, and social relationships. Societal and economic components have become indispensible factors in the future Internet. The transition from the academic Internet to a multifunctional business-involved future Internet puts much higher requirements on the architectural supports to regulate and balance the interests of all stakeholders. In both technical and non-technical aspects, the future Internet architectures are required to provide extensible and flexible explicit interfaces among multiple actors to allow interaction, and enforce policies and even laws. The deep merging of the Internet into everyone’s daily life has made such endeavors and efforts more and more urgent and important. From our perspective, significant research efforts are still needed in aspects such as economics, society, and laws. Experimental facilities: Most of the current testbeds for future Internet architecture research in different countries are results of previous research projects not related to future Internet architectures. The networks use different technologies and have different capabilities. Although the federation efforts are meaningful, they may be restricted in both manageability and capability by such diversity. Testbeds from different countries are also generally tailored or specialized for the architectural design projects of those countries, with different features and emphases. Federation and creating synergy among such testbeds may be challenging. From
our perspective, such challenges also mean a valuable opportunity for research on sharing and virtualization over diverse platforms. Service delivery networks: The key trend driving the growth of the Internet over the last decade is the profusion of services over the Internet. Google, Facebook, YouTube, and similar services form the bulk of Internet traffic. Cloud computing and the proliferation of mobile devices have lead to further growth in services over the Internet. Therefore, Internet 3.0 [16], which is a project in which the authors of this article are involved, includes developing an open and secure service delivery network (SDN) architecture. This will allow telecommunication carriers to offer SDN services that can be used by many application service providers (ASPs). For example, an ASP wanting to use multiple cloud computing centers could use it to set up its own worldwide application-specific network and customize it by a rule-based delegation mechanism. These rules will allow ASPs to share an SDN and achieve the features required for widely distributed services, such as load balancing, fault tolerance, replication, multihoming, mobility, and strong security, customized for their application. One way to summarize this point is that service delivery should form the narrow waist of the Internet (Fig. 1), and content and IP are special cases of service delivery.
IEEE
BEMaGS F
Internet 3.0, which is a project in which the authors are involved, includes developing an open and secure service delivery network architecture. This will allow telecommunication carriers to offer SDN services that can be used by many application service providers.
SUMMARY In this article, we present a survey of the current research efforts on future Internet architectures. It is not meant to be a complete enumeration of all such projects. Instead, we focus on a series of representative research projects. Research programs and efforts from the United States, European Union, and Asia are discussed. By doing this, we hope to draw an approximate overall picture of the up-to-date status in this area.
REFERENCES [1] NSF Future Internet Architecture Project, http://www. nets-fia.net/. ______ [2] NSF NeTS FIND Initiative, http://www.nets-find.net. [3] Named Data Networking Project, http://www.named___________ data.net. _____ [4] MobilityFirst Future Internet Architecture Project, http://mobilityfirst.winlab.rutgers.edu/. [5] NEBULA Project, http://nebula.cis.upenn.edu. [6] eXpressive Internet Architecture Project, http://www.cs.cmu.edu/~xia/. [7] Global Environment for Network Innovations (GENI) Project, http://www.geni.net/. [8] OpenFlow Switch Consortium, http://www.open___________ flowswitch.org/. ________ [9] The FP7 4WARD Project, http://www.4ward-project.eu/. [10] FIRE: Future Internet Research and Experimentation, http://cordis.europa.eu/fp7/ict/fire/ [11] GEANT2 Project, http://www.geant2.net/. _________ [12] AKARI Architecture Design Project, http://akari-project.nict.go.jp/eng/index2.htm ________________ [13] JGN2plus- Advanced Testbed Network for R&D, http://www.jgn.nict.go.jp/english/index.html. [14] China Education and Research Network, http://www.edu.cn/english/. [15] CERNET2 Project, http://www.cernet2.edu.cn/ index_en.htm. _______ [16] Internet 3.0 project, http://www1.cse.wustl.edu/ ~jain/research/index.html. ______________ [17] The Network of the Future Projects of EU FP7, http://cordis.europa.eu/fp7/ict/future-networks/ home_en.html. ________
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
35
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
[18] S. Paul, J. Pan, and R. Jain, “Architectures for the Future Networks and the Next Generation Internet: A Survey,” Comp. Commun., U.K., vol. 34, issue 1, 15 Jan. 2011, pp. 2–42. [19] Future Internet Assembly, http://www.future____________ internet.eu/home/future-internet-assembly.html. _________________________
BIOGRAPHIES ____________ received his B.E. in J IANLI P AN [S] (
[email protected]) 2001 from Nanjing University of Posts and Telecommunications (NUPT), China, and his M.S. in 2004 from Beijing University of Posts and Telecommunications (BUPT), China. He is currently a Ph.D. student in the Department of Computer Science and Engineering at Washington University in Saint Louis, Missouri. His current research is on future Internet architecture and related topics such as routing scalability, mobility, mulithoming, and Internet evolution. His recent research interests also include green building in the networking context.
S UBHARTHI P AUL [S] (
[email protected]) ____________ received his B.S. degree from the University of Delhi, India, and his Master’s degree in software engineering from Jadavpur University, Kolkata, India. He is presently a doctoral
36
Communications IEEE
A
BEMaGS F
student in the Department of Computer Science and Engineering at Washington University. His primary research interests are in the area of future Internet architectures. __________ is a Fellow of ACM, a winRAJ JAIN [F] (
[email protected]) ner of the ACM SIGCOMM Test of Time award and CDACACCS Foundation Award 2009, and ranks among the top 50 in Citeseer’s list of Most Cited Authors in Computer Science. He is currently a professor in the Department of Computer Science and Engineering at Washington University. Previously, he was one of the co-founders of Nayna Networks, Inc., a next-generation telecommunications systems company in San Jose, California. He was a senior consulting eEngineer at Digital Equipment Corporation in Littleton, Massachusetts, and then a professor of computer and information sciences at Ohio State University, Columbus. He is the author of Art of Computer Systems Performance Analysis, which won the 1991 Best-Advanced How-to Book, Systems award from Computer Press Association. His fourth book, High-Performance TCP/IP: Concepts, Issues, and Solutions, was published by Prentice Hall in November 2003. Recently, he co-edited Quality of Service Architectures for Wireless Networks: Performance Metrics and Management, published in April 2010.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Join! www.comsoc.org/NewJoinSpecial
Run Ahead of the Curve
IEEE Communications Society Global Community of Communications Professionals Society Membership benefits include: IEEE Communications Magazine.
Wireless Communication Engineering Technologies (WCET) Certification Program. ComSoc e-News issues. ComSoc Community Directory. member-only products.
___________
Headquarters - New York, USA 3 Park Avenue, 17th Floor New York, NY 10016 USA Tel: +1 212 705 8900 Fax: +1 212 705 8999
[email protected], www.comsoc.org ________
Singapore Office Fanny Su Beh Noi, Manager Solaris, Singapore 138628 Tel. +65 778 2873, Fax: +65 778 9723
[email protected] ______
China Office Ning Hua, Chief Representative Rm 1530, South Twr, Raycom Info Tech Park C. Haidian District Beijing, 100190, China Tel. +86 10 8286 2025, Fax: +86 10 8262 2135
[email protected] ______
______________________
Communications IEEE
______________
______________
Singapore Office Fanny Su Beh Noi, Manager 59E Science Park Drive The Fleming, Singapore Science Park Singapore 118244 SINGAPORE Tel. +65 778 2873, Fax: +65 778 9723
[email protected] _____
_____________________
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
FUTURE INTERNET ARCHITECTURES: DESIGN AND DEPLOYMENT PERSPECTIVES
Loci of Competition for Future Internet Architectures John Chuang, UC Berkeley
ABSTRACT Designing for competition is an important consideration for the design of future Internet architectures. Network architects should systematically consider the loci of competition in any proposed network architecture. To be economically sustainable, network architectures should encourage competition within each locus, anticipate and manage the interactions between the loci, and be adaptable to evolution in the loci. Given the longevity of network architectures relative to network technologies and applications, it is important to ensure that competition is not unnecessarily foreclosed at any particular locus of competition.
DESIGN FOR COMPETITION In contemplating future Internet architectures, the networking community has identified economic viability as a key architectural requirement, along with other requirements such as security, scalability, and manageability. This is in recognition of the fact that any global-scale distributed communications infrastructure requires significant capital investments, and incentives must exist for the network owners to invest in new facilities and services in a sustainable fashion. It is widely accepted that competition is important for promoting the long-term economic viability of the network. This is because competition imposes market discipline on the network operators and service providers, providing them with incentives for continual innovation and investment in their facilities. Conversely, given a lack of competition, a monopoly provider may become complacent and fail to invest in the long-term health of the network. However, competition does not occur automatically in a network. Therefore, “design for competition” must be an important design principle for any future Internet architecture. Designing for competition can also promote the security of a network by encouraging diversity in all levels of the architecture. Robust competition preempts the vulnerabilities of a monoculture [1], including the increased likelihood of correlated failures or cascade failures. Diversity should be sought not just in hardware
38
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
and software levels, but also in the operators of networks, the providers of network services, and the policies for network management and control. As a design principle, design for competition is similar to but different from the “design for choice” principle as articulated by Clark et al. in their influential tussles paper [2]. By designing for choice, Clark et al. refer to the ability of the architecture “to permit different players to express their preferences.” The design for competition principle extends this to the ability of the architecture to permit different players to express their preferences for services by different providers. For example, in thinking about architectural support for user-directed routing, the design for choice principle might suggest that route diversity be made available to end users by a provider over its network. The design for competition principle, on the other hand, might lead to a stronger requirement that the end users can choose routes offered by multiple providers and/or over multiple networks. It should be clear that design for competition does not imply the mandating of competition. Instead, it is an argument against the unnecessary foreclosure of competition. Design for competition means the architecture should be designed to allow for the possibility of supporting multiple providers of a given service, even if the option of having multiple actual providers is not exercised from the start. This is because a network architecture usually outlives the network technologies and network applications. Sometimes, competition may not be feasible or desirable at a particular locus of the architecture. For example, the current generation of network technologies may be subject to strong economies of scale, such that the cost inefficiencies due to competition dominate the benefits obtained from market competition. However, the introduction of new technologies or the adoption of new applications may lead to a reevaluation of the merits of actual competition at that locus at a future point in time. Put another way, a network architecture that is designed for competition would allow innovation, in the form of new entrants and/or new services, to easily be introduced into the network, to compete with existing offerings in the network.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
LOCI OF COMPETITION Designing for competition in a network architecture is not a straightforward exercise. In facilitating a competitive networking environment, one needs to systematically think through the what, where, who (by whom and for whom), when, and how of supporting competition. Network architects can begin by identifying the potential loci of competition in the architecture, that is, candidate locations in a network architecture where a multiplicity of service providers can be supported. Based on the characteristics of prevailing technologies and uses, the architects can also gauge the level of competition that may arise organically for each locus. For example, loci subject to high fixed costs and low marginal costs (i.e., strong economies of scale), strong network effects, or high switching costs should be considered competition-challenged; that is, they tend to face greater obstacles to competition. In the event that a design calls for the preclusion of choice or competition at a particular locus, this decision needs to be clearly justified. In effect, a monopoly is being hard-coded into the architecture. Recognizing that “loci are not silos,” network architects must also anticipate and understand the economic relationships that may exist between the loci of competition. For example, if a firm operates in two vertically related loci (i.e., vertical integration), its market power in one locus may influence the level of competition in the other locus. As another example, two adjacent loci may both be dominated by a small number of firms (i.e., bilateral oligopolies), leading to strategic behavior that may impede innovation on both sides. A critical job of the network architects is to define and develop interfaces to manage and lubricate both the technical and market transactions between the loci.
LOCI PAST, PRESENT, AND FUTURE In the early days of telephony in the United States, with AT&T controlling the long distance lines, the local loops, and even the customer premises equipment (CPE) leased to customers, there was effectively only a single locus of competition. A potential competitor would have had to construct an alternate nationwide network capable of providing end-to-end service in order to compete effectively against AT&T. In reality, strong economies of scale meant it was economically infeasible for such an alternate network to be deployed. Hence, AT&T remained a dominant monopoly, and had to be held in check through regulation for many decades. However, starting in the 1960s and 1970s, supply-side and demand-side changes, in the form of improvements in communication technologies and increasing business demands for telephony services, set the stage for multiple loci of competition to emerge in the telephony network architecture. In particular, the adoption of FCC Part 68 rules in 1975 and the divestiture of AT&T in 1984 made concrete the existence of three separate (but vertically related) loci: CPE, local service, and long distance service. Through explicit intervention by the executive and judicial branches of the government, firms can now com-
pete in each locus individually, and the lowering of entry barriers led to significant innovations in each locus. Even so, the loci remain vertically related, and under the 1996 Telecommunications Act, firms that are dominant in one locus can compete in other loci. Therefore, government oversight remains necessary to this day. For the current Internet architecture, multiple loci of competition can also be identified. In the topological dimension, the architecture includes long distance (backbone) networks, local access networks, and end hosts. In addition, the layering principle on which the Internet is built means that separate loci of competition can also be identified for the physical, network, and application layers of the Internet. Further evolution of the architecture in support of video distribution, cloud computing, and other emerging applications may lead to additional loci of competition at datacenters, content distribution networks (CDNs), authentication authorities, and so on. In contemplating future Internet architectures, it is not unreasonable to expect some loci of competition to look just like those of the current architecture, and some to be entirely different. In addition to devices, conduits, services, applications, and data, even the network architectures themselves can be potential loci of competition. The case has been made, on multiple occasions, that the incumbency of the current IPv4 network architecture has impeded innovation in the network itself [3]. Through network virtualization techniques, it is possible to allow multiple network architectures to be concurrently deployed over a shared physical infrastructure, to compete with one another. Then end users can vote with their feet (or with their data packets) for their favorite, be it IPv4, IPv6, IPvN, or some other network architecture or protocol suite.
IEEE
BEMaGS F
Designing for competition in a network architecture is not a straightforward exercise. In facilitating a competitive networking environment, one needs to systematically think through the what, where, who (by whom and for whom), when, and how of supporting competition.
A 2 × 2 EXAMPLE Network architectures, present and future, can have a large number of loci. As an illustrative example, let us consider a simple network with four loci of competition organized along two dimensions. In the topological dimension, we can separate the network into the edge (E) and the core (C). The edge provides access to end hosts, and is commonly referred to as “the last mile” or “the local loop.” The core provides connectivity between the edge networks over long distances, and is commonly referred to as “the backbone”. In the layering dimension, we can view the network as consisting of a physical layer (P) that carries bits over a physical infrastructure, and a logical layer (L) that sits on top of the physical layer and provides network services to end users. With this simple 2 × 2 matrix (Table 1), we can articulate four different loci of competition. In the physical-edge (P-E) cell of the matrix, competition is manifested in the form of multiple physical connections (wired or wireless) to the end hosts. Telecommunications regulators use the term facilities-based competition to describe competition in this locus. The build-out of new facilities typically involves high up-front costs for trenching, erection of towers, and/or acquisition of radio spectrum. Therefore, the barriers to entry are usually high, and the number of parallel facilities is usually limited. However, once the
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
39
A
BEMaGS F
Communications IEEE
Logical (L)
Physical (P)
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Edge (E)
Core (C)
Number of access service providers
Number and coverage of transit service providers
Number of wired and wireless “last-mile” conduits to each end host
Number and coverage of wide-area physical conduits
Table 1. Loci of competition in a stylized 2 × 2 network architecture. facilities are built, the marginal cost of delivering data is very low, so this part of the industry is said to exhibit strong economies of scale. In the logical-edge (L-E) cell of the matrix, multiple firms may compete in providing network access services to end users. The providers may offer services over different physical networks or over the same physical network. The providers may offer similar services (e.g., IPv4) and compete on quality and/or cost, or may offer different services for different segments of the market (e.g., virtual private networking [VPN] for enterprises). It is not uncommon to find a single firm operating in both the P-E and L-E cells of the matrix: a firm offers network service to end users over a physical infrastructure it owns and operates. In the physical-core (P-C) cell of the matrix, we can have multiple firms who deploy physical infrastructure for wide-area bit transport, such as fiber, submarine cables, terrestrial radio, and satellite links. To the extent that there is overlap in geographic coverage, the firms may compete for business with one another. Finally, in the logical-core (L-C) cell of the matrix, firms might compete with one another for transit customers while at the same time interconnecting with one another to provide global connectivity for their customers. Once again, it is possible, although not necessary, for a firm to be operating in the P-C and L-C cells of the matrix simultaneously. For that matter, it is even possible for a firm to be operating in all four cells of the matrix. The presence of competition at these four loci directly translates into choice available to the end users. First, they may choose among different P-E links based on requirements for mobility, bandwidth, and considerations of cost. End users may even choose to multihome (i.e., connect to multiple physical networks at the same time) for availability or redundancy reasons. The end users can also choose among different L-E providers, with the services offered over different physical networks or over the same physical network. Furthermore, if the network architecture supports some form of userdirected routing, the end users may be able to exercise choice between different L-C and possibly even P-C options. In telephony, for example, consumers can select long-distance carriers on a per-call basis today by dialing additional prefixes. In sum, the end user can, in theory, exercise (or delegate) choice in all four cells of the matrix.
LOCI ARE NOT SILOS While we can identify multiple loci of competition in a network architecture, it would be a mistake to think that we can design for competition one locus at a time, ignoring any interactions
40
Communications IEEE
A
BEMaGS F
between the loci. In reality, there are many different ways through which competition in one locus can affect that in another locus.
SUBSTITUTES First, consumer choice in one locus may sometimes serve as a substitute for choice in another locus. For example, if the network architecture seeks to promote route diversity, it could do so through the support of multi-homing (first-hop diversity), and/or through the support of userdirected routing (ith-hop diversity). Multihoming requires P-E (and possibly L-E) diversity, while user-directed routing is often targeted at L-C (and possibly P-C) diversity. To the extent that end users can access meaningful routing alternatives, multihoming and user-directed routing are imperfect substitutes, so a network architect can weigh the relative costs and benefits of supporting one or both mechanisms in his/her architectural design, and justify his/her design accordingly. P-E competition (facilities-based competition) and L-E competition (e.g., through open access regulation) is another good example of potential substitutes in the architecture. In the 1990s, when facilities-based competition was considered weak in the United States, regulators and legislators pushed for L-E competition by requiring the physical network operators to open up access to their facilities (e.g., through unbundled network elements) to competing service providers. As a result, competitive local exchange carriers (CLECs) were able to offer alternative L-E service to consumers by gaining access to the existing local loop from the incumbent local exchange carriers (ILECs) such as the Baby Bells, rather than build out their own physical network. In response, the ILECs put up non-technical obstacles to access their facilities (e.g., switches in central offices, cell towers) in their efforts to frustrate competition in the L-E locus. In retrospect, one can view the CLEC experiment as a failure: open access may relieve the pressure for facilities-based competition, but it is not an effective substitute for P-E competition, as a monopoly P-E operator can still tilt the playing field in the L-E locus.
VERTICAL INTEGRATION A network architecture should not preclude a single firm from operating in more than one locus of the competition matrix. A firm may be vertically integrated by operating in both the L-E and P-E loci (e.g., Comcast), in both the L-C and P-C loci (e.g., Level 3), or even in all four loci (e.g., AT&T). Once vertical integration comes into play, however, a network architect needs to be cognizant of the competition dynamics that arise. The benefit of vertical integration derives from the fact that an integrated firm (e.g., one operating in both L-E and P-E) may provide service to consumers more cheaply than two independent firms operating in L-E and P-E separately. This economies of scope cost savings may come from a variety of sources (e.g., joint facilities operation, billing, customer service). For example, in Fig. 1b, the vertically integrated provider may be able to provide the same service at a lower cost, and therefore offer it at a lower price, than two separate providers at the logical and physical layers.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
To the extent that economies of scope are strong, vertical integration is socially desirable, and the industry structure may be dominated by vertically integrated firms, as illustrated in Fig. 1c. From a consumer’s perspective, there is still effective choice in providers. However, this industry structure may discourage innovation at one of the loci. For example, a new firm with an innovation in the L-E locus may not be able to offer the new service unless it also enters the PE market, which may have a high entry barrier. Furthermore, vertical integration can be detrimental to competition, and therefore social welfare, if it allows a firm to use its dominant position (i.e., market power) in one locus to compete unfairly in another locus. Figure 1d offers an example where a vertically integrated firm enjoys a monopoly position in the P-E market, and has the ability to exercise its monopoly power to frustrate competition in the L-E market. In our earlier example, the Baby Bells (e.g., Pacific Bell, Bell Atlantic) were vertically integrated ILECs that held monopoly positions in their respective P-E markets, and were accused of unfairly competing against the CLECs in the L-E market. In an even earlier example, AT&T prior to the 1984 divestiture was vertically integrated across the local and long-distance telephony markets, and was accused of using its monopoly power in the local telephony market to compete unfairly against MCI, Sprint, and other competitors in the long-distance telephony market. Even after the competing long-distance carriers successfully sued to gain access to interconnection with AT&T’s local networks, AT&T was still able to tilt the playing field in its favor by effecting hidden subsidies from its local service to its longdistance service. It eventually took anti-trust action by the U.S. Department of Justice to force a “vertical disintegration” of AT&T, in 1984, into separate firms for local and long-distance service. The story continues in the 1990s and 2000s, when the digitization of voice and video, together with the widespread availability of mobile data networks, meant that local telephony operators no longer had a monopoly in the P-E market. Consequently, anti-trust objections to vertical integration no longer hold, and vertical re-integration occurred. Given the historical evolution of the industry, it should be clear that the current market structure should not be assumed as the final equilibrium state of the industry. While multiple competitors occupy the P-E market today, continuing technological advances and/or changing consumer requirements may once again shift this locus toward a monopolistic market in the future.
TYING The designation of AT&T as the exclusive carrier in the United States for the iconic Apple iPhone when it was first launched in 2007 is a notable recent example of tying. By selling one product or service only in conjunction with another product or service, tying can lead to competition dynamics in two loci. Similar to vertical integration, tying is not a major concern if consumers have choices in alternative products and services in both loci, or if potential entrants do not face high entry barriers into either locus. In the case of iPhone/AT&T, consumers have
(a)
(c)
(b)
(d)
IEEE
BEMaGS F
Figure 1. Examples of vertical industry structures: a) no vertical integration — users choose logical layer (green) and physical layer (white) providers separately; b) with vertical integration, users can either choose logical and physical layer providers separately, or choose an integrated provider; c) all firms vertically integrated — users choose one integrated provider; there is a higher entry barrier for new entrants; d) competition at the logical layer with monopoly at the physical layer — the user has no choice in physical layer service; a vertically integrated firm has control over whether or how the user has choice in logical layer service.
ample alternatives for both device and carrier, so tying is not considered problematic from an antitrust perspective. On the other hand, should one or both of the tying firms enjoy sufficient market power in their loci, the practice may in fact be enjoined by antitrust laws (e.g., the Sherman Antitrust Act and the Clayton Act in the United States). For the 700 MHz spectrum auction in the United States in 2008, the FCC adopted “open device” and “open application” rules for the highly sought after “C” Block. The rules effectively preempt Verizon, the eventual auction winner, from engaging in tying of devices or applications to services offered over its 700 MHz network. Given the ability of the winner to build out a high-quality nationwide network using the spectrum, the open rules were believed to encourage greater innovation in the loci of both mobile devices and applications.
DELEGATION, GATEKEEPING, AND NETWORK NEUTRALITY Both design for choice and design for competition principles stipulate that users be permitted to express their preferences for services. In a general sense, this may be realized by an interface that allows users to make selections in each of the loci of competition in the architecture. Some users or applications will clearly benefit from the ability to exercise full control over the selection of services in all loci. However, it is not clear that, even in a simple 2 × 2 architecture, most or many users will actually wish to make separate, explicit selections for each of the loci. Instead, it is more realistic to expect a typical user to: • Select a service provider in one of the loci, relying on the service provider to select services in the other loci
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
41
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
U
G
U U
U
(a) B B
G
(c) B
U G
(b)
(d)
Figure 2. Examples of delegation and gatekeeping. Selections made by user, broker, and gatekeeper are labeled U, B, and G, respectively: a) mix and match — user selects logical (green) and physical (white) layer services; b) delegation — user delegates choice to third party broker; c) physical gatekeeper — user selects physical layer service, which in turn selects logical layer services; d) logical gatekeeper — user selects logical layer service, which in turn selects physical layer services. • Delegate the service selection decisions to a third-party broker, who will compose a service package out of options in the various loci to match the preferences of the user Figure 2 illustrates, for a simple two-layer network architecture example, how delegation may be differently realized. Figure 2a shows, in the absence of delegation, how a user will have to explicitly select providers in both the physical and logical layers. In Fig. 2b, the user selects a third-party broker, and delegates the selection of physical and logical services to the broker. In Fig. 2c, the user selects a physical layer provider, and relies on the latter to select logical layer services on its behalf. In this case, the physical layer provider serves as the gatekeeper to logical layer services. Finally, in Fig. 2d, the user selects a logical layer provider, and relies on the latter to select physical layer services on its behalf. The logical layer provider serves as the gatekeeper to the physical layer services. Given the complexity and dynamism of large distributed networks, service providers or thirdparty brokers are naturally in a much better position to collect and act on low-level network information to make service composition decisions on behalf of their users. However, is there any difference, from a competition or innovation perspective, between the various forms of delegation? The answer is yes. If there is a relative difference in strength of competition in the two loci, allowing the firms in the less competitive locus to serve as gatekeepers controlling access to services in the other locus will result in reduced competition and innovation in both loci [4]. Typically, competition is weaker in the physical layer because of the necessity of significant up-front capital investments. In this case, the physical gatekeeper model of delegation in Fig. 2c would be the least desirable from an economic perspective. Architecturally, this means that it is important for the interface to allow end users or third party brokers to explicitly select logical layer services, independent of the physical layer provider. This will prevent a physical gatekeeper from becoming
42
Communications IEEE
A
BEMaGS F
an obstacle between innovative logical layer services and the users who desire them. Architectural proposals like ROSE (Routing as a Service) [5], CABO [6], and Cabernet [7] offer possible paths for the network to move away from a physical gatekeeper model, which currently dominates the Internet, to a model based on logical gatekeepers and/or delegation to third-party brokers. Generalizing to more than two loci, we should ensure that the network architecture does not allow a competition-challenged locus to assume a gatekeeping role over any other locus. This is entirely consistent with, and perhaps provides a fundamental economic argument for, the principle of network neutrality. If the physical access network locus is the competition-challenged locus, then we must prevent the dominant firms in this locus from having the ability to discriminate against different offerings in the applications, services, content, and devices loci. In the absence of explicit prohibition, architecturally or otherwise, we should expect firms in the competition-challenged loci to exercise their market power over service selection and become de facto gatekeepers. At the same time, these firms have no incentives whatsoever to embrace changes in the architecture that may shift the power of service selection to other loci. Therefore, a network architecture cannot afford to be silent on this issue. One possible design choice is to foreclose, at the architectural level, the possibility of gatekeeping by the competition-challenged loci. This leaves open the possibility of gatekeeping by other more competitive loci. However, if we are uncertain of the long-run relative competitiveness of different loci in the architecture, an alternate approach may be to explicitly disallow gatekeeping by any locus, and to rely on delegation to independent third-party brokers.
EVOLUTION Network architectures evolve over time, and so do the loci of competition. The evolution can occur in several ways. First, the level of competition in a given locus may change over time. Consolidation via horizontal mergers may change a competitive locus to an oligopolistic or even a monopolistic one. Conversely, new entrants with novel technologies or business models may introduce competition to a locus. As we have seen in the preceding discussions, these changes can affect levels of competition in adjacent loci as well. Second, new technologies and services may create new loci of competition that did not previously exist and/or destroy old ones. In the process, the adjacency of loci may also be rearranged. In the context of the current Internet architecture, we have seen an emergence of content distribution networks (CDNs) and large content providers (LCPs, e.g., Google) as major sources of interdomain traffic [8]. We can consider the provision of these services as new loci of competition. Furthermore, the datacenters for these CDN/LCP servers are being deployed with direct connections to the consumer access networks, bypassing the backbone networks altogether. Consequently, the vertical relationship between the oligopolistic local access locus and the com-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
petitive long-distance locus, traditionally subject to scrutiny by telecom economists and regulators, may be supplanted by the bilateral oligopolistic relationship between the local access and the CDN/LCP loci. Traditionally, bilateral oligopolies are characterized by long-term contracts negotiated between players with strong market power. This has important implications to innovation in both of the loci. The “paid peering” agreements that are currently being negotiated between local access networks and large content providers are precisely the type of long-term contracts that will shape the industry for years to come. At the same time, new loci may also create new opportunities for vertical expansion by existing firms and redefine tussle boundaries. For example, backbone operator Level 3 expanded into the CDN locus and began to deliver video streams for customers like Netflix. This led to a dispute between Level 3 and Comcast in late 2010 regarding compensation for Comcast’s carriage of Netflix traffic. At issue is whether Comcast and Level 3 are two Tier-1 networks with settlement-free peering agreements, or if Level 3 is a CDN terminating traffic within the access networks of Comcast. This ongoing dispute highlights the challenge in characterizing the business relationship between two firms when both are vertically integrated across different but overlapping loci of competition.
MINIMIZE SWITCHING COSTS In addition to ensuring that a multiplicity of providers can be supported at each locus, there is another important role to be played by the architect. It is to influence how often, and how easily, the choice of providers can be exercised at each locus. Specifically, the lower the cost of switching providers, the easier it is for service consumers to try out new innovative services, and the more competitive a locus can be. In the search engine market, for example, Hal Varian contends that Google has every incentive to continue to innovate because “competition is only one click away” [9]. So, should a network architecture allow choices to be made on a per-year basis, per-application basis, or per-packet basis? The appropriate timescale of choice will likely differ for each locus, depending on the nature of service in question and the amount of overhead incurred by the service provider each time a customer adds or drops the service. Extra attention should be given to those loci that are inherently more competition-challenged (e.g., due to strong scale economies). For example, number portability proved to be a particularly effective way to reduce switching costs in the local access locus of the telephony network architecture. Developments in software-defined radio technologies, as another example, offer the prospect of dramatically lowering switching costs for the wireless local access locus. Importantly, the objective should be to minimize switching costs on both the customer side and the provider side. If we focus on only the former but ignore the latter, the provider will have greater incentive to manage its own incurred costs by discouraging switching behavior altogether. In particular, the provider can
create an artificial switching cost to the customers by employing long-term service contracts with early termination penalties. Such a strategy of contractual lock-in may serve the provider’s goal of churn reduction, but it can also dampen competition significantly.
IEEE
BEMaGS
TAKEAWAYS Let us close with a summary of main takeaways from this article for network architects who are contemplating the design of future Internet architectures. First, the architecture should ensure that a multiplicity of providers can be supported at each locus. We do not want to unnecessarily foreclose competition anywhere in the network. Second, we need to recognize that architecture outlasts technologies and applications. The level of competition in each locus may change over time, and we cannot anticipate or dictate the number of choices in each locus. Third, in choosing designs that facilitate competition at each locus, we want to pay particular attention to those loci that are naturally competition-challenged due to stronger economies of scale, etc. In general, minimizing switching costs can be an effective way to promote competition in any locus. Finally, remembering that “loci are not silos,” we cannot design for competition one locus at a time. Instead, the network architecture must facilitate robust competition in face of a wide range of possible strategic interactions between providers in different loci, as well as providers that straddle multiple loci. Interfaces must be carefully developed to manage and lubricate both the technical and market transactions between the loci.
F
Remembering that “loci are not silos,” we cannot design for competition one locus at a time. Instead, the network architecture must facilitate robust competition in the face of a wide range of possible strategic interactions between providers in different loci, as well as providers that straddle multiple loci.
REFERENCES [1] D. Geer et al., “Cyberinsecurity: The Cost of Monopoly,” Comp. and Commun. Industry Assn. Report, 2003. [2] D. Clark et al., “Tussles in Cyberspace: Defining Tomorrow’s Internet,” IEEE/ACM Trans. Net., vol. 13, no. 3, June 2005, pp. 462–75. [3] T. Anderson et al., “Overcoming the Internet Impasse Through Virtualization,” Computer, vol. 38, no. 4, Apr. 2005, pp. 31–41. [4] P. Laskowski and J. Chuang, “Innovations and Upgrades in Virtualized Network Architectures,” Proc. Wksp. Economics of Networks, Systems, and Computation, 2010. [5] K. Lakshminarayanan, I. Stoica, and S. Shenker, “Routing as a Service,” UC Berkeley EECS tech. rep. UCB/CSD04-1327, 2004. [6] N. Feamster, L. Gao, and J. Rexford, “How to Lease the Internet in Your Spare Time,” ACM Computer Commun. Rev., Jan. 2006. [7] Y. Zhu et al., “Cabernet: Connectivity Architecture for Better Network Services,” Proc. Wksp. Rearchitecting the Internet, 2008. [8] C. Labovitz et al., “Internet Inter-Domain Traffic,” Proc. ACM SIGCOMM, 2010. [9] H. Varian, “The Economics of Internet Search,” Angelo Costa Lecture delivered in Rome, Feb. 2007.
BIOGRAPHY JOHN CHUANG (
[email protected]) ________________ is a professor in the School of Information at the University of California at Berkeley, with an affiliate appointment in the Department of Electrical Engineering and Computer Science. His research interests are in economics of network architectures, economics of information security, incentives for peer-to-peer systems, and ICT for development. He received his Ph.D. in engineering and public policy from Carnegie Mellon University, his M.S.E.E. from Stanford University, and graduated summa cum laude in electrical engineering from the University of Southern California.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
43
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
FUTURE INTERNET ARCHITECTURES: DESIGN AND DEPLOYMENT PERSPECTIVES
Biological Principles for Future Internet Architecture Design Sasitharan Balasubramaniam, Waterford Institute of Technology Kenji Leibnitz, National Institute of Information and Communications Technology and Osaka University Pietro Lio’, University of Cambridge Dmitri Botvich, Waterford Institute of Technology Masayuki Murata, Osaka University
ABSTRACT Currently, a large number of activities on Internet redesign are being discussed in the research community. While today’s Internet was initially planned as a datagram-oriented communication network among research facilities, it has grown and evolved to accommodate unexpected diversity in services and applications. For the future Internet this trend is anticipated to continue even more. Such developments demand that the architecture of the new-generation Internet be designed in a dynamic, modular, and adaptive way. Features like these can often be observed in biological processes that serve as inspiration for designing new cooperative architectural concepts. Our contribution in this article is twofold. First, unlike previous discussions on biologically inspired network control mechanisms, we do not limit ourselves to a single method, but consider ecosystems and coexisting environments of entities that can cooperate based on biological principles. Second, we illustrate our grand view by not only taking inspiration from biology in the design process, but also sketching a possible way to implement biologically driven control in a future Internet architecture.
INTRODUCTION The Internet has transformed and changed our lives in many aspects over recent years, where the increase in popular and sophisticated services continues to attract users. However, the current Internet infrastructure, which was built mainly for conservative data traffic usage, is approaching its limit. This has led the research community to investigate solutions toward the future Internet within various project initiatives (e.g., GENI, FIND, AKARI, FIRE) [1]. The motivation for this research development is largely driven by the behavioral changes and
44
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
needs of users toward the Internet. While the original Internet was designed mainly to allow users to exchange information (mostly to support research and work), today we see highly diverse sets of services, many of which are used daily to enhance our quality of life. These services range from information gathering mechanisms tailored to our personal and societal needs,to support for various social problems, as well as entertainment. In order to fully support such diverse services, the future Internet will require new architectural and protocol designs. This new architectural design would need to integrate highly intelligent processes to improve their robustness, scalability, efficiency, and reliability. This is particularly crucial because the number of devices in the future is expected to drastically increase. One approach to provide this capability that communications researchers have recently started investigating is through bio-inspired processes [2–4]. Biologically inspired mechanisms have been applied in recent years to diverse types of networks (e.g., sensors, wireless, fixed, services). However, from the viewpoint of the future Internet, current bio-inspired approaches are only a first step toward realizing a fully functional system. The main reason behind this is because most of these bio-inspired solutions have only tackled specific problems for a particular type of network. In order to realize the full potential of bio-inspired solutions for the future Internet, these disparate solutions need to be designed to function in a fully integrated manner. In this article, we aim at paving the way for this vision to become a reality. We first summarize some of the key requirements of the future Internet, in particular from the core network infrastructure perspective. This is then followed by discussions on relevant properties found in biological processes that enable multiple organisms and systems to coexist in an ecosystem,
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
where our aim is to combine various bio-inspired network control mechanisms within the future Internet. Our proposal is built on two intrinsic properties of biological systems, which includes a fully integrated system of systems within an organism, as well as organisms that can coexist in an ecosystem. The article illustrates how our idea could be augmented with some existing proposed architectures for the future Internet. Lastly, we present our grand vision of future communication networks that are directly driven by biological systems, taking the concept of biologically inspired networking to a new level.
REQUIREMENTS OF THE FUTURE INTERNET While the future Internet will cover various different types of agendas, we only focus in this article on the core network infrastructure of the Internet. In the future, we can anticipate greater heterogeneity, coexistence, and cooperation among different types of networks. This can range from different content and service distribution networks to virtual networks that all operate over the same physical network. At the same time, communication networks of the future will require emergent properties embedded directly into the networks. This is often dubbed as self-* properties in the literature (self-organization self-management, etc.). In this section we list the requirements expected in our view of the future Internet.
VIRTUALIZATION AND ADAPTIVE RESOURCE MANAGEMENT A crucial component of communication networks is resource management, and its efficient usage will determine the quality of service (QoS) delivered to end users. Before the Internet gained its current popularity, single network providers usually owned the communication infrastructures. However, this situation is slowly transforming into a new business model, where a distinction between network (infrastructure) providers and service providers is becoming apparent. This is usually referred to as virtual networks, where service providers lease the resources they need from network providers and are allowed to have a certain control over usage of these resources. The increased flexibility means that service providers may configure their virtual network according to the services they are offering, while the network provider needs to safeguard the fair usage of the network. However, the dynamics of services may change over short timescales, leading to the need for dynamic resource subscription policies from the network provider. Another crucial requirement of the future Internet is the adaptive usage of resources through efficient routing. One important research agenda is the need for scalable, robust, and distributed routing applicable to large-scale networks. The majority of current routing solutions are based on optimization methods, where prior knowledge of traffic demand exists, and the demand does not change frequently. Performing routing this way is ideal if reconfigura-
tions are only required over long timescales. However, as the number of services increases and evolves at a fast pace, more reactive and intelligent routing mechanisms are required.
IEEE
BEMaGS
ENERGY EFFICIENCY As the Internet’s popularity increases, so has its supporting information and communications technology (ICT) infrastructure. This ranges from increases in data centers to host services, network access technologies, as well as end-user devices. Overall, this has led to a steadily increasing consumption of energy to operate the infrastructure. As the traffic volume in the future is anticipated to increase, this in turn will also lead to a higher amount of energy consumption by networking equipment. A common approach toward saving energy today is switching devices off or putting them into sleep state. However, with the large number of nodes anticipated in the future Internet, this process should be performed in a collaborative manner, while ensuring that end users’ requirements are met.
F
In the future, we can anticipate greater heterogeneity, coexistence, and cooperation among different types of networks. This can range from different content and service distribution networks to virtual networks that all operate over the same physical network.
FLEXIBLE AND EVOLVABLE INFRASTRUCTURE A major factor behind the requirement of redesigning the Internet is the fact that the original Internet was designed mainly for accommodating data traffic with stable traffic patterns. However, it is neither feasible nor practical to perform a complete redesign of the Internet each time new requirements, or drastic technological or social changes arise that do not fit the current architecture. Therefore, the design of the future Internet should include a sustainable infrastructure that is able to support evolvability. This should enable new protocols to be introduced with minimal conflict to existing ones. At the same time, the design of architectures and protocols should be made in a modular way, where protocol components can have cross-layer interactions. The evolvability of the future Internet should also allow for a certain degree of openness, where protocols with the same functionalities can be deployed by various entities to suit their own needs; but these protocols must be able to coexist with each other and minimize any possible conflicts. For example, different service providers should be able to deploy their own routing algorithm that best suits their customers’ QoS/quality of experience (QoE) requirements.
SERVICE-ORIENTED PROVISIONING A key point that has attracted users to the Internet is the continual development and provisioning of new and more advanced services (rich multimedia content). It is expected that this trend will also continue in the future. Therefore, as a multitude of new services start to flood into the Internet, it is essential that these services are autonomous and capable of exhibiting self-* properties, similar to the requirements for network devices of the future. These properties should allow services to autonomously discover and combine with other services in an efficient and distributed manner. At the same time, deployment of these services should not be restricted to end systems, but may also be embedded into network routers. This section discussed some of the require-
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
45
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Over billions of years, Communication system
evolved to suit the changes of the environment, and for this very reason
Internal regulatory system
Communication system
Sensory system
adaptability has
Sensory system
this resilience and
Cognitive system
Molecular scale
bio-inspired
Organism scale
techniques have provided inspiration
Internal regulatory system
Society
Society
Ecosystem
Figure 1. Hierarchical ecosystem in a biological system.
for communication network researchers.
ments for designing a sustainable future Internet that is able to handle current and new challenges. Its main features are that the network should support virtual networks and allow each virtual network to adaptively subscribe resources from underlying networks, have self-* properties for managing itself, enable energy efficiency, and support a diverse set of services in a flexible way. Similar characteristics can often be found in biology, and the following section discusses some biological processes that may serve as inspiration to meet these requirements.
BASIC MECHANISMS OF BIOLOGICAL COOPERATION Biological systems have remarkable capabilities of resilience and adaptability. These capabilities are found in various biological organisms, ranging from microorganisms to flocks of animals and even human society. Over billions of years, this resilience and adaptability has evolved to suit the changes of the environment, and for this very reason bio-inspired techniques have provided inspiration for communication network researchers [2–4]. In particular, there are two especially appealing aspects of biological systems that could be beneficial in designing architectures of the future Internet. First, biological systems are always composed of a multitude of protocols that combine various processes to control different elements of an organism. Second, biological systems as a whole exhibit a hierarchical ecosystem structure that allows various organisms and systems to coexist. Figure 1 illustrates an example of both these aspects, presenting an abstract layered view of internal functionalities within organisms, composed of an internal regulatory system, a cognitive system, a sensory system, and a communication system. In the remainder of this section we describe some key features of this abstract model and its importance in allowing biological systems to coexist in the manner in which they do.
INTERNAL REGULATORY SYSTEMS In order for biological systems to maintain stability and survive through age, there is a need for self-regulation to balance the system and maintain constant conditions in the face of external and internal perturbations. One example of
46
Communications IEEE
this self-regulation process found in organisms is homeostasis. There are a number of different homeostasis processes ranging from thermoregulation to blood glucose regulation. Homeostasis requires the integration of information from different parts of the body, as well as the analysis and forecast of resources. Another example of an internal regulatory system within an organism is the immune system, which is able to fend off non-self invader cells. These biological mechanisms can serve as good inspiration in the design of self-management and self-regulation in communication networks since they operate efficiently, and are robust without centralized control.
BIOLOGICAL SENSORY SYSTEMS Organisms possess a number of sophisticated sensory systems to maintain internal balance. These systems receive their external inputs from sensors and propagate the stimulus through a complex hierarchical network to various components within an organism. Examples of sensory systems are the central and peripheral nervous systems in the human body. Another example is the lateral line, which is a sensing organ found in aquatic organisms. Sensory systems interconnected through the nervous system or lateral line provide a medium for coordinating and transmitting signals between various parts of the body. Insight on where to locate processing units within a communication network can be gained from observing the structure of the nervous system (e.g., the location of ganglia serving as hubs between the peripheral and central nervous systems).
BIOLOGICAL COMMUNICATION AND SIGNALING A key property of biological systems is the ability for entities to communicate and signal between each other. This form of signaling can come in various forms, ranging from speech to chemical signaling. Signaling is required for synchronization between organisms. At the microorganism level, reaction-diffusion describes the concept of morphogenesis, the process where chemicals are released and diffused between cells during tissue development to explain patterns of stripes or spots on animal coats. Another example is quorum sensing, which is when cells signal among each other and cooperate in the face of environmental changes. All of these processes are cooperative, and different entities communicate in a
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
fully distributed way among each other. In most cases the organism is unaware of the emergent outcome of synchronization in the whole system. Societies are formed between various organisms through such signaling and communication processes. They are the very foundation that allows organisms to function collectively and exhibit various self-* mechanisms in order to coordinate various tasks. During migration, flocks of birds use signaling between members in the pack to rotate the leader of the flock to balance energy expenditure of each bird due to wind resistance. Social insects, such as ants and bees, show a high degree of cooperation, and perform division of labor and self-organization while foraging [5].
INTERACTING POPULATION DYNAMICS Population models, such as predator-prey interactions, competition, and symbiosis, describe the interactions between different species while coexisting within a common space or ecosystem. In predator-prey, the predators are the dominant of two interacting species and feed on the prey. On the other hand, symbiosis occurs when both species coexist and mutually benefit in their growth, while under competition both populations mutually inhibit each other. The population dynamics in the ecosystem determines whether the system is able to maintain balance among competing species. Attaining balanced coexistence among heterogeneous populations (networks, services) in a common ecosystem is one of the goals of our proposal.
FUTURE INTERNET ARCHITECTURE In this section we discuss how we could possibly map the biological mechanisms of the previous section to example architectures that have been proposed for the core infrastructure of the future Internet. These architectures are the Services Integration, Control, and Optimization (SILO) architecture [1] and Information Transfer Data Services (ITDS) [1], although other architectures can also be treated in a similar way. Our aim is to use an ecosystem model, as shown in Fig. 1, as a basis to ensure that biological processes can simultaneously coexist with the other processes that are involved in its environment.
EXTENSION OF THE SILO ARCHITECTURE The aim of the SILO architecture is to create a modularized architecture for the future Internet that can: • Create building blocks from fine-grained services • Allow these building blocks to be combined in order to support complex communication tasks • Allow cross-layer interactions between different services Based on these aspects, the aim is to allow the application layer to select the most appropriate services to support its needs. A positive aspect of the SILO architecture is its ability to create modular architectures based on the capabilities and resources of the end devices. For example, in resource constrained device such as sensors, only vital services are embedded directly
into the device to ensure minimum energy consumption, while other services are loaded on demand. For these reasons, SILO is very suitable for our proposed bio-inspired future Internet architecture, where each biological process can represent a SILO service. Similar to the SILO solution, the processes invoked by the biological mechanisms depend on rules and constraints that govern the relationship between the processes. Therefore, each biological process will have a description that includes: 1. The specific function it performs (e.g., routing) 2. The key requirements of those functionalities from external parameters (e.g., measurements) 3. The interfaces that are compatible with other processes In the case of 1, this allows applications to determine the appropriate biological process that meets its requirements in the deployed environment. Such characteristics include overhead of the protocols and delay incurred when the process is applied to a specific topology. At the same time, parameters are defined for each of these processes to maintain behavioral constraints, which could be applied through policies. The SILO architecture currently uses a control agent that determines the services to be composed together. In a similar fashion, a control agent can determine the most appropriate biological process to support the type of application. Figure 2 illustrates an example of the bioinspired SILO architecture and its application to a virtual network above a physical network. The figure also illustrates the protocol stack and the different processes fitting into the stack. The paths along the virtual network are selected using a noise-driven internal regulatory mechanism known as attractor-selection, an internal regulation system found in E. coli cells [6]. The underlying network uses the reaction-diffusion mechanism for signaling between the nodes, and the routing process is based on chemotaxis [7], which is a motility mechanism used by microorganisms to attract a gradient found in an environment. Also similar to the original SILO services, each process is equipped with “knobs” to allow external tuning of parameters. The layered protocol shows how the different processes in the routing and overlay path layers interact with each other. The attractor-selection mechanism is defined through a number of states and is driven by noise. Internal regulation changes the state due to external influences; that is, this internal regulation controls the bandwidth resource for the virtual path. Once a certain threshold is exceeded, attractor-selection interacts with the underlying network, which uses chemotaxis to discover new routes. The chemotaxis mechanism is a distributed routing mechanism that selects the path node by node from the source to the destination following the highest gradient. The gradient is formed through the node-to-node interaction of the reaction-diffusion process [7]. An example of the interaction between two bio-inspired control mechanisms is shown in Fig. 3. The figure shows how paths are discovered
IEEE
BEMaGS F
The population dynamics in the ecosystem determines whether the system is able to maintain balance among competing species. Attaining balanced coexistence among heterogeneous populations (networks, services) in a common ecosystem is one of the goals of our proposal.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
47
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Application layer
Overlay network
Attractor selection
Attractor Attractor state 1 state 2
Attractor state n
Objective Knobs
Noise-driven selection
Attractor-selection
Interface
Measurement
QoS, SRC, DST Chemotaxis Reaction diffusion
Route discovery Knobs
Underlay network
Node load
Link load
Hop count Chemotaxis
Interface
Signaling information Route discovery
Knobs
Reaction-diffusion
Physical layer
Figure 2. Bio-inspired SILO architecture.
through the gradients using the chemotaxis model for both the primary and secondary paths. Initially, attractor-selection determines the overlay path 1 (red line) to be chosen. Since the QoS requirements are fulfilled, the system remains in a stable state and is kept at the dynamic attractor for path 1 despite small fluctuations. At time step t = 400 congestion occurs, leading to the path becoming unstable, which in turn decreases the QoS of the overlay layer. The path is then switched from path 1 to path 2 (blue line), which offers greater stability, and at time step t = 600 the system becomes stable again. This simple example shows how different dynamic biologically inspired control schemes can symbiotically cooperate in reacting to congestion and changes of traffic conditions at various timescales. Coming back to our ecosystem model presented in Fig. 1, we can see that the different biological mechanisms are applied at the molecular level (attractor-selection and chemotaxis), and are able to coexist within an organism. Figure 4 shows how the concept of modularity and openness can be realized within the bioinspired future Internet. Two routing mechanisms (ant-based routing [5] and chemotaxis) have been applied to the underlying networks. Each routing algorithm consumes a certain quantity of resources, but both mechanisms can symbiotically coexist in the network, supporting the requirement of openness. The most ideal routing mechanism may depend on different objectives (e.g., scalability, timeliness for route discovery, reaction to dynamics, energy of signaling overhead). Therefore, through the symbiotic requirements, various new protocols can be updated and added.
EXTENSION OF THE ITDS ARCHITECTURE Another example is the bio-inspired ITDS, which focuses on service provisioning as a key requirement. The aim of ITDS is to have transfer of
48
Communications IEEE
information in the underlying network rather than just raw data as is currently done. This proposal is realized through sets of services that are embedded into the routers to perform application-based processing. This is far from the traditional approach, which only allows end hosts to have service intelligence while the underlying network is used for forwarding packets. Whereas the original ITDS only handles the application layer requirements, we extend this by allowing embedded processes into the network layer as well. This in turn allows the network to be highly adaptive and support evolving services. An example of bio-inspired ITDS and its corresponding protocol stack are shown in Fig. 5. In this figure we show a combination of two services on two virtualized planes, the security and multimedia planes. On the security plane we have a reliability function that works in sequence with the privacy function, where the reliability function is based on an immune system mechanism. At the multimedia plane, we consider a codec service that works in conjunction with a caching service that can extract data to serve various end users. We assume that each of these services is able to migrate from node to node. Since one of the objectives of the future Internet is energy efficiency, we also have an embedded service that measures energy output within a node. Cooperative signaling is also performed between the nodes to permit certain nodes to be switched off while others take the burden of the traffic to minimize overall energy consumption. Cooperative signaling is performed using the quorum sensing process, and the energy efficiency service is based on the internal thermoregulation process of an organism. When we map this back to the ecosystem model of Fig. 1, we can see that this example expands further from the example used for bio-inspired SILO. In this example, our bio-inspired processes in the underlying network are based on mechanisms found at
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Primary path
Secondary path
Path 1
Path 1
Path 2 Path 2
Destination
Chemotaxis gradient over links
Congestion
1.2
1.2
1
1
0.8
0.8
Path priority
Attractor selection switches primary path
Path priority
Source
0.6 0.4 0.2 0
0.6 0.4 0.2
0
100 200 300 400 500 600 Time steps
0 600
700
800 900 Time steps
1000
Figure 3. Illustration of cooperative adaptation between attractor-selection and chemotaxis gradient-based routing.
the molecular level, while the processes used to manage the security and multimedia plane, as well as their interactions, are based on internal regulatory system of an organism. So the example shows how processes at the microorganism level can coexist with processes at the organism level, representing a virtual network operating over a physical network.
BIOLOGICALLY-DRIVEN FUTURE NETWORKS While biologically inspired mechanisms have provided increased capability and adaptability for communication networks, a major challenge is understanding a biological process and developing an appropriate algorithm. At the same time, most bio-inspired algorithm designers use
refined biological processes that omit various hidden functionalities, where these functionalities may solve foreseeable future problems. Therefore, a viable alternative to the current approaches is to allow systems to be directly driven by biological systems — or as we term it biologically driven future networks — bypassing the step of using artificial bio-inspired algorithms. An example of our proposed concept is illustrated in Fig. 6. In this example a cell may represent a virtualized overlay network, and in the event of changes in the overlay network environment, this triggers feedback into the cell culture. This feedback may lead to mitosis, where the output from the mitosis process can be filtered back to the overlay to reconfigure the overlay network into two virtual networks. This vision may be one possible solution toward the development of network devices for the future
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
49
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Homeostasis Chemotaxis
Application layer
Shortest path
Homeostasis
Chemotaxis
Antrouting
Symbiosis
Symbiosis
Physical layer Chemotaxis
Antrouting
Figure 4. Illustration of evolvable protocol for SILO architecture.
no requirement for defining a complex protocol and designing countermeasures for all possible kinds of communication network scenarios. However, there are still a number of challenges before such an approach could finally be realized. First, biological systems require a favorable environment in terms of nutrients and temperature to be cultivated, cells may die faster than they reproduce, and they may react differently or uncontrollably, all of which must be catered for. It may also lead to new security challenges. At the same time, synchronization between the biological organism and operations within the physical network will be a challenge, since biological processes at microorganism level can take hours to show some effects. New software and hardware design will also be required, where hardware modules must be able to house the biological culture, and the network conditions must be fed back to the biological environment.
Internet, particularly to cope with increased and unknown complexities. As shown in the figure, a biological culture of microorganisms could directly drive the behavior of the underlying network, and when a change is experienced in the physical network, a feedback process could manually be injected to change the environment of the biological culture. From a practical perspective, we need to limit this vision to the use of microorganisms. In essence, this allows us to harness and directly exploit communication processes at the nano/molecular scale of biological systems [8] to control communication systems. Directly using biological systems for various systems has been investigated previously. For example, in [9] slime mould was used to design the Greater Tokyo railway network, while in [10] slime mould was again used to design the U.S. road networks. However, we believe that a similar concept could also be extended to the realtime management of communication networks. Through this approach, a new methodology of tackling problems in future networks can be devised with the concept of biological software/ hardware co-design (where biological cultures could be interfaced to software and hardware systems). A major benefit would be that there is
CONCLUSION In this article we discuss approaches to support the design of the future Internet architecture by making use of biological mechanisms. A
Multimedia caching
Video codec
Multimedia plane Autonomic nervous system
Video codec
Reliability
Reliability Privacy
Thermoregulations
Caching Privacy Immune system
Autonomic nervous system Security plane
Chemotaxis Quorum sensing
Network layer
Network layer Chemotaxis Quorum sensing
Network layer
Link layer
Link layer
Link layer
Physical layer
Physical layer Thermoregulation
Physical layer
Figure 5. Bio-inspired ITDS.
50
Communications IEEE
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Culture of cells, microorganisms
Interface between living organism and communication networks
Feedback
Virtualized network 1
Virtual overlay
Virtualized network 2
Underlying network
Figure 6. Biologically driven future networks. large number of different bio-inspired methods have previously been proposed for enabling communication networks to exhibit self-* capabilities, where the majority of these solutions has only focused on individual specific mechanisms to solve a particular problem. However, from the future Internet perspective, we need to take a step back to see how these existing bio-inspired solutions can be integrated to meet the many requirements of the future Internet. In this article we outline possible solutions of integrating biologically inspired processes into the future Internet architecture to support this need. Specifically, we illustrate how this could be augmented with two existing architectures proposed for the future Internet, SILO, and ITDS. While applying biologically inspired methods may improve the robustness, adaptability, and evolvability of a new Internet design, our grand vision is communication networks of the future that can be directly driven by biological systems.
ACKNOWLEDGMENT The authors would like to thank Tokuko Haraguchi and Yuji Chikashige of NICT, Japan, for permitting usage of the HeLa cell images. The authors wish to acknowledge the following funding support: Science Foundation Ireland via the “Federated, Autonomic Management of End-to-End Communications Services” (grant no. 08/SRC/I1403), Science Foundation Ireland via “A Biologically Inspired Framework Supporting Network Management for the Future Internet” (grant no. 09/SIRG/I1643), and EU FP7 grant “RECOGNITION: Relevance and Cognition for Self-Awareness in a Content-Centric Internet.”
REFERENCES [1] S. Paul, J. Pan, and R. Jain, “Architectures for the Future Networks and the Next Generation Internet: A Survey,” Comp. Commun., vol. 34, no. 1, 15 Jan. 2011, pp. 2–42. [2] M. Meisel, V. Pappas, and L. Zhang, “A Taxonomy of Biologically Inspired Research in Computer Networking,” Comp. Networks, vol. 54, no. 6, Apr. 2010, pp. 901–16. [3] F. Dressler and O. B. Akan, “Bio-Inspired Networking: From Theory to Practice,” IEEE Commun. Mag., vol. 48, no. 11, Nov. 2010, pp. 176–83. [4] P. Lio’ and D. Verma, “Biologically Inspired Networking,” IEEE Network, vol. 24, no. 3, May/June 2010, p. 4. [5] G. Di Caro and M. Dorigo, “AntNet: Distributed Stigermetic Control for Communication Networks,” J. Artificial Intelligence Research, vol. 9, 1998, pp. 317–65. [6] K. Leibnitz, N. Wakamiya, and M. Murata, “BiologicallyInspired Self-Adaptive Multi-Path Routing in Overlay Networks,” Commun. ACM, vol. 49, no. 3, Mar. 2006, pp. 62–67. [7] S. Balasubramaniam et al., “Parameterised Gradient Based Routing for the Future Internet,” Proc. IEEE Advanced Information Networking and Application (AINA), Bradford, U.K., 2009. [8] I. F. Akyildiz, F. Brunetti, and C. Blazquez, “Nanonetworks: A New Communication Paradigm,” Computer Networks, vol. 52, June 2008, pp. 2260–79. [9] A. Tero et al., “Rules for Biologically Inspired Adaptive Network Design,” Science, vol. 327, no. 5964, Jan. 2010, pp. 439–42. [10] A. Adamatzky, “Physarum Machines: Computers from Slime Mould,” World Scientific Series on Nonlinear Science, Series A, 2010.
BIOGRAPHIES SASITHARAN BALASUBRAMANIAM (
[email protected]) ________ received his Bachelor (electrical and electronic engineering) and Ph.D. degrees from the University of Queensland in 1998 and 2005, respectively, and Master’s (computer and communication engineering) degree in 1999 from Queensland University of Technology. He joined the Telecommunication Software and Systems Group (TSSG), Waterford Institute of Technology, Ireland, right after completion of his Ph.D.. He is currently the manager for the Bio-Inspired Network research unit at the TSSG. He has worked on a number of Irish funded (e.g., Science Foundation Ireland, PRTLI) and
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
51
A
BEMaGS F
Communications IEEE
52
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
EU projects. His research interests include the bio-inspired future Internet as well as molecular communications.
diseases dynamics, and bio-inspired communications and technology.
KENJI LEIBNITZ (
[email protected]) _____________ received his Master and Ph.D. degrees in information science from the University of Würzburg, Germany, where he was also a research fellow at the Institute of Computer Science. In May 2004, he joined Osaka University, Japan, as a Postdoctoral Researcher and from 2006 until March 2010 a Specially Appointed Associate Professor at the Graduate School of Information Science and Technology. Since April he is a senior researcher at the Brain ICT Laboratory of the National Institute of Information and Communications Technology in Kobe, Japan, and an invited associate professor at Osaka University. His research interests are in modeling and performance analysis of communication networks, especially the application of biologically inspired mechanisms to self-organization in future networks.
DMITRI BOTVICH (
[email protected]) __________ received his Bachelor’s and Ph.D. degrees in mathematics from Moscow State University, Faculty of Mechanics and Mathematics, Russia, in 1980 and 1984, respectively. He is currently the chief scientist of the Scientific and Technical Board at the Telecommunication Software and Systems Group, Waterford Institute of Technology. He currently leads the PRTLI FutureComm project at the TSSG, and has coordinated and worked in a number of EU and Science Foundation Ireland projects. He has published over 100 papers in conferences and journals, and currently supervises seven Ph.D. students. His research interests include bio-inspired autonomic network management, security, trust management, wireless networking, queuing theory, optimization methods, and mathematical physics.
P IETRO L IO ’ (
[email protected]) __________ is a senior lecturer and a member of the Artificial Intelligence Division of the Computer Laboratory, University of Cambridge. He has an interdisciplinary approach to research and teaching due to the fact that he holds a Ph.D. in complex systems and nonlinear dynamics (School of Informatics, Department of Engineering of the University of Firenze, Italy) and a Ph.D. in (theoretical) genetics (University of Pavia, Italy). He supervises six Ph.D. students and teaches the following courses: Bioinformatics (Computer Laboratory), Modeling in System Biology (System Biology Tripos, Department of Biochemistry), Models and Methods in Genomics (MPHIL Comp Biology — Department of Mathematics), and 4G1 System Biology (Department of Engineering). His main interests are in investigating relationships and effectiveness of different biosystems modeling methodologies (particularly multiscale approaches), modeling infectious
MASAYUKI MURATA [M] (
[email protected]) ______________ received M.E. and D.E. degrees in information and computer science from Osaka University, Japan, in 1984 and 1988, respectively. In April 1984 he joined the Tokyo Research Laboratory, IBM Japan, as a researcher. From September 1987 to January 1989 he was an assistant professor with the Computation Center, Osaka University. In February 1989 he moved to the Department of Information and Computer Sciences, Faculty of Engineering Science, Osaka University. In April he became a professor at the Cybermedia Center, Osaka University, and is now with the Graduate School of Information Science and Technology, Osaka University since April 2004. He has more than 400 papers in international and domestic journals and conferences. His research interests include computer communication networks, performance modeling, and evaluation. He is a member of ACM and IEICE.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
CONTINUING EDUCATION FOR COMMUNICATIONS PROFESSIONALS
- Increase your problem-solving skills - Review and expand on what you already know - Quickly become familiar with the latest advancements, regulations, & standards - Learn from experts who focus on practical application - Earn CEUs to meet your professional development requirement
COMSOC TRAINING EVENTS Wireless Communications Engineering - Current Practice Virtual One-Day Course Instructor: Javan Erfanian Wed, July 20, 2011 - 9:00am - 4:30pm EDT
LTE for the Wireless Engineering Practitioner: Fundamentals & Applications Virtual One-Day Course Instructor: Luis Blanco Wed, August 3, 2011 - 9:00am - 4:30pm EDT
To learn more, visit comsoc.org/training comsoc.org/training
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
FUTURE INTERNET ARCHITECTURES: DESIGN AND DEPLOYMENT PERSPECTIVES
Enabling Future Internet Research: The FEDERICA Case Peter Szegedi, Trans-European Research and Education Networking Association Jordi Ferrer Riera and Joan A. García-Espín, Fundació i2CAT Markus Hidell, Peter Sjödin, and Pehr Söderman, KTH Royal Institute of Technology Marco Ruffini and Donal O’Mahony, Trinity College Dublin Andrea Bianco and Luca Giraudo, Politecnico di Torino Miguel Ponce de Leon and Gemma Power, Waterford Institute of Technology Cristina Cervelló-Pastor, Universitat Politècnica de Catalunya Víctor López, Universidad Autónoma de Madrid Susanne Naegele-Jackson, Friedrich-Alexander University of Erlangen-Nuremberg
54
Communications IEEE
ABSTRACT
INTRODUCTION
The Internet, undoubtedly, is the most influential technical invention of the 20th century that affects and constantly changes all aspects of our day-to-day lives nowadays. Although it is hard to predict its long-term consequences, the potential future of the Internet definitely relies on future Internet research. Prior to every development and deployment project, an extensive and comprehensive research study must be performed in order to design, model, analyze, and evaluate all impacts of the new initiative on the existing environment. Taking the ever-growing size of the Internet and the increasing complexity of novel Internet-based applications and services into account, the evaluation and validation of new ideas cannot be effectively carried out over local test beds and small experimental networks. The gap which exists between the small-scale pilots in academic and research test beds and the realsize validations and actual deployments in production networks can be bridged by using virtual infrastructures. FEDERICA is one of the facilities, based on virtualization capabilities in both network and computing resources, which creates custom-made virtual environments and makes them available for Future Internet Researchers. This article provides a comprehensive overview of the state-of-the-art research projects that have been using the virtual infrastructure slices of FEDERICA in order to validate their research concepts, even when they are disruptive to the test bed’s infrastructure, to obtain results in realistic network environments.
The European Community co-funded project FEDERICA (Federated E-Infrastructure Dedicated to European Researchers Innovating in Computing Network Architectures) [1] supports the development of the future Internet by definition. The project consortium is an optimal mixture of research institutes, universities, and industrial partners in order to foster cross-sector collaboration. Although, the two-and-a-half-year project officially ended on 30 October 2010, after a four-month extension, the infrastructure and its services are still up and running thanks to the voluntary efforts of the European Research and Education Networking (NREN) organizations, which are actively participating in the operation of the infrastructure. FEDERICA’s architecture has been designed in such a way that allows users to request “virtual slices” of the infrastructure that are completely isolated from each other and behave exactly as if they were physical infrastructure from the point of view of the users’ experiments. The major objective of the deployment of such an infrastructure is twofold. On one hand, FEDERICA allows extensive research on virtualization based architecture itself. Experiments on various resource virtualization platforms, virtual resource descriptions, monitoring and management procedures, virtual network slice compositions, isolations or federations can be performed. On the other hand, the virtual slices of FEDERICA enable multidisciplinary research on future Internet within the slices. This article is organized as follows. After a brief summary of the FEDERICA virtualization
0163-6804/11/$25.00 © 2011 IEEE
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
capable infrastructure and its services, we provide an overview of similar initiatives around the world emphasizing the importance of federated facilities. Then, we discuss the special features of FEDERICA that make the infrastructure unique (as of today) by way of supporting a wide variety of future Internet research activities, maybe leading to new paradigms. As FEDERICA is highly user-centric, we introduce the broad spectrum of its user community by looking at the actual users and use cases. The contribution of FEDERICA user experiments to the whole Internet research space is also illustrated in this article that finally concludes with the project’s perspective for the future.
Infrastructure 1 Gb/s circuits Core links Non-core
IEEE
F
KTH
HEANet PSNC
DFN
CESNET HUNGARNET
SWITCH SARR RedIRIS
I2CAT
FCCN GRNET ICCS
Figure 1. FEDERICA infrastructure, the substrate topology. tions can be preselected by users). This may eventually lead to a complete portfolio of cloud services, where all of these services share simplicity in access, typically through a web interface, simple scalability, and ease of use.
FEDERATED FACILITIES WORLDWIDE FEDERICA does not operate in isolation. There are initiatives all over the world, which are of particular relevance and represent the natural set of peers for liaison and collaboration with FEDERICA. United States — The Global Environment for Network Innovation (GENI) initiative [3] in the United States is an activity that aims at creating a unique virtual laboratory for at-scale networking experimentation. GENI is in its third phase, called spiral 3, of exploratory rapid prototyping with about 50 experiments and facilities operational. Work in spiral 3 should improve the integration, operations, tools, and documentation of GENI prototypes to lower the barriers of use. User projects have to collectively agree on a federation architecture to create the GENI environment spanning multiple facilities. The FEDERICA project has obtained excellent interactions with the GENI project office and some GENI experiments (e.g., ProtoGENI, GpENI) through regular meetings. Japan — The AKARI Architecture Design Project in Japan [4] aims to implement a new generation network by 2015, developing a network architecture and creating a network design based on that architecture. The philosophy is to pursue an ideal solution by researching new net-
IEEE Communications Magazine • July 2011
Communications
BEMaGS
NORDUNET
FEDERICA AS A SERVICE FEDERICA project participants have designed and deployed a virtualization capable infrastructure substrate, including programmable highend routers, multi-protocol switches, and PC-based virtualization capable nodes, on a pan-European footprint (Fig. 1). The physical network topology is composed of 13 sites, 4 core and 9 non-core nodes, connected by 19 point-to-point links. On top of this substrate, virtual infrastructure slices can be created, which realize whatever topologies are requested by the users. The facility is also connected to the public Internet, thereby enabling easy access from any location and type of connectivity (wireless or fixed line) [1]. Virtualization is the key having a profound influence in technology. In its network-oriented meaning, it corresponds to the creation of multiple virtual networks on top of a physical infrastructure. The network virtualization can be performed at most layers of the ISO/OSI stack, from the data link to the network layer. In its system-oriented sense the process is even faster and more intriguing, leading to multiple operating systems with varying tasks running on the same hardware platform. The service architecture of FEDERICA follows the Infrastructure as a Service (IaaS) paradigm. IaaS, in principle, is the common delivery of hardware (e.g., server, storage, network), and associated software (e.g., operating systems virtualization technology, file systems) as a service. It is an evolution of traditional hosting that does not require any long term commitment and allows users to provision resources on demand. Amazon Web Services Elastic Compute Cloud (EC2) and Secure Storage Service (S3) are examples of commercial IaaS offerings [2]. FEDERICA has two major services: • The provisioning of best effort or QoS IP slices with the option of both preconfigured and unconfigured resources (i.e., routing protocols, operation systems, QoS parameters) • The provisioning of raw resources in terms of data pipes (i.e., native Ethernet links) and unconfigured resources (i.e., empty virtual machines, clear routing tables, no network protocols) The current FEDERICA services may naturally evolve toward Platform as a Service (where various preconfigured images and network scenarios can be provided from a repository), and Application as a Service (where a set of applica-
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
55
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
45% Based on the submitted slice requests Based on the potential user community
40% 35% 30% 25% 20% 15% 10% 5% 0% Configured resources, IP best-effort slice
Configured resources, IP QoS slice
Not configured resources, IP connectivity
Raw resources and “data pipes”
Figure 2. User segmentation results on the exploited unique features of FEDERICA. work architectures from a clean slate without being impeded by existing constraints. Some prototyping projects have been started at the University of Tokyo in collaboration with the National Institute of Information and Communications Technology (NICT). Europe — In Europe, the FIRE initiative [5] from the European Commission funds various infrastructure projects (e.g., Panlab, Wisebed, OneLab) with which FEDERICA already has strong collaboration. The federation of OneLab and FEDERICA facilities is an excellent example in terms of combining guaranteed and nonguaranteed resources. FEDERICA can offer resources to OneLab that have full access to the network layers and OneLab can provide access to a larger set of distributed computing resources. During the last years, OneLab has developed an architecture to federate different domains based on PlanetLab that is called Slice Federation Architecture (SFA) [6]. SFA enables FEDERICA and similar infrastructures to become a worldwide federated facility supporting collective on future Internet research.
UNIQUE FEATURES OF FEDERICA FEDERICA can be easily compared with PlanetLab [7]. PlanetLab started in Princeton, New Jersey, and is a well-known global research facility based on a large set of end hosts on top of the public Internet. The hosts are running a common software package that includes a Linuxbased operating system. The key objective of this software package is to support distributed virtualization (i.e., the ability to allocate a slice of PlanetLab network-wide hardware resources to a user application). In contrary, a FEDERICA node is not only considered as an end host as in PlanetLab, but a combined network and computing element that itself routes traffic and affects network reachability. In addition, the user is not forced to use a specific operating system or
56
Communications IEEE
A
BEMaGS F
application but it is free to install any software component within the slice. While PlanetLab connections are based on the public Internet, FEDERICA allows the user to perform experiments on the lower layers as well. FEDERICA architecture is more than an evolution of the concept of a traditional test bed. Test beds are usually devoted to a few technologies and oriented to production service or to pure research, and are not flexible enough to accommodate both types of use. The FEDERICA environment is also different from a commercial cloud environment, where the delay between nodes is typically negligible. In FEDERICA the delay corresponds to physical distance of the substrate nodes. Such naturally distributed architecture provides an ideal environment for testing medium size experiments and migration paths in a realistic environment. The unique features of FEDERICA (as of today) can be summarized as follows: •FEDERICA has its own physical substrate fully controlled and managed by the FEDERICA Network Operation Center. This allows the user to take over control of the lower layer resources within their slices. At the moment, only raw Ethernet resources are available for the users but in the near future native optical resources (i.e., lambdas) may also be provided. •Thanks to having full control over the substrate links, specific QoS parameters can be assured for the virtual connection of the user slices and real transmission delays can be experienced. Currently the links run up to 1 Gb/s but this may be upgraded radically in the near future. •FEDERICA users have full flexibility and freedom to choose any networking protocol or operating system to be installed on their virtual nodes. To the virtualization platform end, JUNOSbased programmable high-end Juniper platforms and VMware-based SUN servers are both currently deployed in the FEDERICA substrate. •FEDERICA can ensure the reproducibility of the complete testing environment and conditions of the user experiments at a different location or time. Repeatability of the experiments can also be ensured in the sense of obtaining the same results given the same initial conditions at any time. •The overall architecture is federation-ready in line with the Slice Federation Architecture concept. Moreover, FEDERICA provides, for example, non-web-based federated SSH access to all of its resources supported by the SingleSign-On infrastructure of the research and education community. However, it is important to mention some of the limitations of FEDERICA, too. Typically, compared to PlanetLab or commercial clouds, the scalability of the FEDERICA virtual infrastructure is limited. The scalability is the function of the given size of the physical substrate (that may be extended in the future) and the number of active user experiments requiring QoS assurances on links and/or computing nodes. As a consequence of this, user access to the FEDERICA slices must be governed with care. This role is undertaken by the User Policy Board of the project, which collects and approves (or rejects) user slice applications.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
The exploitation of the aforementioned unique features of FEDERICA is illustrated in this article by a wide range of user projects.
70%
A
BEMaGS F
Based on the submitted slice requests Based on the potential user community
60%
USERS AND USE CASES During the active lifetime of the FEDERICA project (2008–2010) the consortium partners approached and consulted more than 40 potential user groups all over Europe that resulted 15 slice requests for specific user experiments, as of 30 October, 2010. The segmentation and analysis of the user community have been considered to be important to FEDERICA. The aim of this segmentation was to understand the user community better and the type of experiments that can exploit the unique features of FEDERICA. This should help to steer the future infrastructure development and user consultation activities in the right direction so as to support the needs of the community better.
USER SEGMENTATION The detailed results of the FEDERICA user segmentation are available in the public project deliverables [8]. In the following section, we highlight the main characteristics of the user community. Analyzing both the slice requests already submitted and the preliminary requirements collected from the possible user groups, Fig. 2 shows that 40 percent of the performed experiments in FEDERICA use a slice as a fully configured IP network with best effort connections. There is nothing unique in that sense, so other motivations (e.g., cost, economy of scale, simplicity of use) must exist in the case of that 40 percent. In contrast, 60 percent of the experiments performed use FEDERICA because of its unique features (e.g., IP quality of service assurance or unconfigured resource provisioning). In particular, 26.6 percent of users have requested raw resources and data pipes (i.e., IaaS) in order to fully configure and manage their slices. Analyzing the potential broader user base, we expect a few more user experiments in the future exploiting the above mentioned unique features of FEDERICA.
MAJOR EXPERIMENTS FEDERICA’s mission is to support the development of the future Internet via Future Internet Research projects. To measure its success, FEDERICA has collected feedback from users in order to understand, say, the contribution of the experiments’ results to the ICT research community as a whole. Figure 3 shows the results of the users’ feedback on this particular aspect. Sixty percent of the experiments performed in FEDERICA contribute to international research projects partly funded by the European Commission. More than 10 percent of experiments are aimed at directly or indirectly contributing to standardization activities in the field of networking. Requests from the commercial or private sector has not yet been received during the active lifetime of the project, but some discussions within the community suggest that it could be possible in the future.
50% 40% 30% 20% 10% 0% National project support
International/ EC project support
Standardisation/ open source development support
Commercial/ private research support
Figure 3. Contribution of FEDERICA experiments to ICT research community.
CONTRIBUTION TO FUTURE INTERNET RESEARCH In the following section, we illustrate the usage of FEDERICA via a wide spectrum of its user experiments. The examples of user experiments are grouped in three categories according to their main objectives; validation of virtual infrastructure features, evaluation of multilayer network architectures, and design of novel data and control plane protocols.
VALIDATION OF VIRTUAL INFRASTRUCTURE FEATURES This group of experiments aims at validating the basic principles of virtualization capable infrastructures in general and particularly the unique features of FEDERICA. The Universitat Politècnica de Catalunya (UPC), Spain, requested a FEDERICA slice, called ISOLDA, in order to perform basic network performance and slice isolation tests. Parameters such as bandwidth, latency, jitter, and packet loss were measured in parallel slices with the aim to prove sufficient isolation between them. Figure 4 depicts the test scenario where two virtual connections (red and dashed green slices) share a physical port on the FEDERICA substrate node in the middle. This shared physical link contains two different VLANs, one for each virtual connection. VM1 and VM3 are located on the same virtual machine server, called VMServer 1, while VM2 and VM4 are located on another server, called VMServer 2. The other virtual connection (dotted blue line) does not share any resource. The isolation test at network level was done by using various control and management protocols to identify whether the isolated virtual machines can accidentally connect to each another. Isolation of virtual machines was also proven by ping tests. The results from these tests were affirmative, as the virtual machines on different slices were not visible to one another. In conclusion, information was obtained about how the FEDERICA substrate nodes should be configured to assure the isolation feature between
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
57
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
VMServer 1 eth0
VMServer 2 eth0
VM1 eth1 eth0 VM3
VM2 eth1 eth0 VM4
Figure 4. ISOLDA: Network performance and slice isolation tests. slices that share physical resources. Following these defined procedures in FEDERICA, the isolation between slices is completely ensured. KTH Royal Institute of Technology, Sweden, requested a slice, called METER, in order to study another important feature of FEDERICA: the repeatability of its experiments. This user project dealt with problems related to repeated experiments on a shared network, where other external activities may influence the set of results. KTH investigated a method to identify time periods of comparable network conditions based on metadata-contextual information about the environment where the experiment was executed. During the timeframe of an experiment, active background measurements were run to collect metadata parameters. Experiment data was time-stamped and KTH used statistical analysis on the metadata to determine if experiment data was gathered during comparable network conditions. An important goal was to be able to do this without interrogating the experiment data itself. Within the slice, service degradation in terms of resource competition with experimental activities in other slices was rarely detected. An important part of the experimental work was to detect differences in network conditions e.g., in terms of latency. A set of background measurements was used to identify periods of comparable network conditions. Based on the measurement results, it was possible to reduce the confidence intervals of the outcome of the repeated experiment, so the remaining values were suitable for comparison. The Friedrich-Alexander University of Erlangen-Nuremberg (FAU), Germany, in cooperation with FEDERICA partner Deutsche Forschungsnetz (DFN), requested a slice in order to perform Hades Active Delay Evaluation System (HADES) measurements over the physical infrastructure of FEDERICA. The nodes of the FEDERICA substrate were measured in a full mesh topology; with bidirectional measurements, this yielded 42 measured links. HADES was developed and deployed as a tool that provides performance measurements and offers IP performance metrics such as oneway delay, one-way delay variation, and packet loss [9]. The main purpose of the experiment was to collect and archive these IP performance metrics over an extended period of time and make the data available to other FEDERICA project partners and experiments that needed reference data as part of their investigations on the behavior of virtualized networks and slice processing.
58
Communications IEEE
A
BEMaGS F
The data was also used to show that FEDERICA users had a stable network environment and repeatable conditions for their experiments.
EVALUATION OF MULTI-LAYER NETWORK ARCHITECTURES This group of experiments illustrates the capability of connecting external test beds or virtual slices provided by other facilities to the FEDERICA user slice. In this way, multi-layer and multi-domain network architectures and federation principles can be evaluated. Trinity College Dublin (TCD), Ireland, requested a slice, called OIS, in order to carry out experiments on the Optical IP Switching (OIS) concept, a network architecture [10] that was developed by the CTVR Telecommunications Research Centre based at the University of Dublin. The idea behind OIS is to build a packet router that can analyze traffic flows and use an underlying optical cross-connect to bypass the router electronics for suitable IP flows. The FEDERICA slice was used to analyze how dynamic creation, modification or cancellation of optical paths can degrade the quality of applications transported over TCP protocols, on a large-scale testbed implementation. The results presented substantial differences to those previously obtained through laboratory-based experiments. The OIS network architecture was setup in TCD’s laboratory as a core network and interconnected with vanilla IP domains, which were realized within the FEDERICA slice. In the experiment the FEDERICA slice was organized in such a way that emulated a simple access network, where user data were aggregated by an IP router and sent toward the TCD testbed. Here packets were routed back to FEDERICA towards the destination network. TCD could then verify how their testbed would work when connected to external legacy IP networks. The results showed, for example, that UDP and TCP transmissions are impaired when the signals are dynamically switched from an electronic to an optical connection if both upstream and downstream links are congested. The experiment also gave TCD the opportunity to evaluate feasibility and efforts required to use virtualized networks in combination with physical testbeds. The Telecommunications Software & Systems Group (TSSG), based in Waterford, Ireland, was one of the 10 partners of the European Commission funded project PERIMETER [11] which is finished in May 2011. The project addressed challenges such as quality of experience (QoE), security, and end user impact in order to establish a new paradigm for seamless mobility. TSSG requested a FEDERICA slice (incorporating five virtual nodes and routers) to extend the existing physical test bed and carry out more robust testing of the PERIMETER application particularly in the areas of routing, scalability, and authentication. This was a relatively inexpensive way to dynamically pool and use resources for experimentation of new networking concepts. The federated PERIMETER testbed and FEDERICA slice were used to conduct a number of testing processes including scenario conformance tests, interoperability tests, application tests, Living
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Labs tests, and performance and scalability tests in the project. These testing processes could not be performed while achieving such a level of realism without the use of FEDERICA. In conclusion, it was validated that the federation of a physical test bed and a virtual slice allows experimental testing to proceed, using real-size platforms, real operating systems, and real applications in a realistic environment, which is similar to the actual target system. The Universidad Autónoma de Madrid (UAM), in Spain also requested a FEDERICA slice, called ANFORA, in order to interconnect that with two other infrastructures. This complex networking experiment used resources of three facilities: FEDERICA, OneLab, and PASITO. Beside the FEDERICA slice, UAM already had OneLab nodes with specific monitoring cards in order to perform quality of service measurement and was also connected to PASITO, the Spanish national layer 2 networking facility [12], to provide lower-layer connections at the FEDERICA points of presence. The multilayer test facility that was created (Fig. 5) was used to evaluate the Path Computation Element (PCE) protocol, and to validate multilayer traffic engineering (MTE) algorithms. The FEDERICA slice allowed the creation of a virtual network topology, which played the role of an IP service provider’s backbone network. OneLab measurement tools, implemented on top of FEDERICA virtual nodes, ensured precise monitoring of traffic engineering (TE) parameters, like bandwidth or delay. The end-to-end monitoring information was sent to the PCE database so, PCE nodes knew the actual performance of each layer. During the experiment, end-user traffic was routed using the IP layer (i.e., FEDERICA slice) by default, while the network performance was adequate. However, when there was IP congestion, the lower layer connections of PASITO infrastructure were used to provide the desired quality of service for the end users. PCE computed the end-to-end routes and decided on which network layer the traffic should be transmitted. As a result of the experiment it was concluded that, PCE is suitable for multilayer algorithms since their computation is more complex than traditional routing algorithms. Moreover, the PCE database was filled with information of both layers providing a global view of the network to the operator. The interconnection of FEDERICA, OneLab, and PASITO provided a unique test environment in order to successfully validate various MTE algorithms.
DESIGN OF NOVEL DATA AND CONTROL PLANE PROTOCOLS This group of experiments aims at designing and validating novel data and control plane protocols as well as architectures. The Politecnico di Torino (PoliTo), Italy, requested a slice, called VMSR, in order to experiment with the Virtual Multistage Software Router (VMSR) distributed architecture. Distributed routers are logical devices whose functionalities are distributed on multiple internal elements, running on virtual machines, in order to achieve larger aggregated throughput and improved reliability. The VMSR multistage router architecture [13] was composed by three
UMA premises
IEEE
BEMaGS F
UPNA premises One-lab FEDERICA
End user
End user
PASITO
Figure 5. ANFORA: investigation of federation and multilayer network scenarios. internal stages: layer-2 load balancers, Ethernet switches to interconnect components, and standard Linux based software routers. In order to coordinate the internal elements and to allow VMSR to behave externally as a single router, custom-made control and management protocols were designed by PoliTo. The VMSR experiment in FEDERICA consisted of three L2 balancers (Linux virtual machines running Click on Ethernet Virtual Switch), one Ethernet switch, three L3 routers (i.e., standard PCs running the Linux protocol stack and XORP), and three external nodes that generate and sink traffic. The main advantage of implementing the VMSR experiment in FEDERICA was related to the ability of controlling and monitoring network resources allocation inside the architecture, which would not be possible in a testbed (e.g., on top of the public Internet). Functional tests on the control and management plane protocols were successfully performed. PoliTo measured throughput and latencies on all the involved links while varying the packet size of traffic generators to determine the suitability of the infrastructure supporting high-performance applications. Last but not least, Fundació i2CAT, the Spanish research foundation, requested a slice in order to perform scalability tests of the Harmony multi-domain and multi-vendor network resource brokering system developed by the European Commission funded project PHOSPHORUS [14]. Harmony defines the architecture for a network service layer between the grid middleware and applications and the network resource provisioning systems (NRPS). Harmony architecture has evolved from a centralized network service plane (NSP) model to a distributed NSP model, passing through a middle stage, the multilevel hierarchical NSP model, as depicted in Fig. 6. Moreover, hybrid architectures of the NSP have also been under consideration for huge high-capacity environments. The experiment performed over the FEDERICA slice aimed to analyze the performance and scalability of the different Harmony NSP topologies under different workloads. The experiment focused on collecting information about two parameters: the service provisioning time and the average rate of blocking requests. The slice
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
59
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
The PCE database was filled with information of both layers providing a global view of the network to the operator. The interconnection of
Centralized
FEDERICA, OneLab
Hierarchical Basic architectures
Daisy chain
and PASITO provided a unique test environment in order to successfully validate various MTE algorithms. Meshed
Hybrid Advanced architectures
Figure 6. PHOSPHORUS: Harmony network resource brokering system scalability study. provisioned by FEDERICA consisted of computational resources (15 virtual machines with 2 Gbytes RAM and 10 Gbytes storage each), where the Harmony NSP entities were deployed. The performed test cases produced measurements that helped i2CAT to better understand and characterize the performance and scalability of the developed system under higher loads than the ones used in the physical test bed available in the PHOSPHORUS project. Several new results were obtained, such as the comparison of the performance of each of the feasible architectures of the NSP. Moreover, the PHOSPHORUS project participants obtained service provisioning times of very different situations and loads for each test scenario through reconfiguring the FEDERICA slice.
CONCLUSION In this article we have given a brief overview of the unique features and services of the FEDERICA virtualization capable infrastructure, developed and deployed in the framework of a European Commission funded project. The broad spectrum of Future Internet Research experiments, performed over the virtual slices of FEDERICA, is also illustrated above. Based on the feedback of the user community, it is clearly signaled that the FEDERICA facility is essential for sustainable support of Future Internet Research in Europe and beyond. That is why the European NREN partners of the former project consortium have agreed to continuously support the operation of the infrastructure and are willing to do further development under the recently launched Network Factory task of the GN3 project’s Joint Research Activity 2 [15].
ACKNOWLEDGMENT The FP7 project FEDERICA was partially supported by the European Commission under Grant Agreement no. RI-213107. The authors
60
Communications IEEE
acknowledge the contribution of all project partners, user projects, and specifically for their input, Mauro Campanella (GARR, project coordinator of FEDERICA), Peter Kaufmann (DFN), Vasilis Maglaris (NTUA), Frances Cleary, and Eileen Dillon (TSSG).
REFERENCES [1] P. Szegedi et al., “With Evolution for Revolution: Managing FEDERICA for Future Internet Research,” IEEE Commun. Mag., vol. 47, no. 7, July 2009, pp. 34–39. [2] S. Bhardwaj, L. Jain, and S. Jain, “Cloud Computing: A Study of Infrastructure of a Service (IaaS),” Int’l. J. Eng. and Info. Tech., vol. 2, no. 1, Feb. 2010, pp. 60–63. [3] J. B. Evans and D. E. Ackers, “Overview of GENI Overview of GENI and Future Internet in the US,” 22 May 2007, http://www.geni.net/ [4] H. Harai, “AKARI Architecture Design Project,” 2nd Japan EU Symp. New-Generation Network and Future Internet, Oct. 13, 2009, http://www.akari-project.jp/ [5] FIRE WHITE PAPER, 2009, “FIRE: Future Internet Research and Experimentation,” Aug. 2009, http://www.ict-fire.eu/home/publications/papers.html [6] S. Soltesz et al., “On the Design and Evolution of an Architecture for Federation,” ROADS2007, July 2007, Warsaw, Poland. [7] L. Peterson, “PlanetLab: A Blueprint for Introducing Disruptive Technology into the Internet,” Jan. 2004, http://www.planet-lab.org/. [8] FP7-FEDERICA, Deliverable DNA2.3 “FEDERICA Usage Reports,” Nov. 2010, http://www.fp7-federica.eu/. [9] P. Holleczek et al., “Statistical Characteristics of Active IP One Way Delay Measurements,” Int’l. Conf. Net. and Services (ICNS’06), Silicon Valley, CA, 16–18 July 2006. [10] M. Ruffini, D. O’Mahony, and L. Doyle, “Optical IP Switching: A Flow-Based Approach to Distributed Cross-Layer Provisioning,” IEEE/OSA J. Opt. Commun. and Net., vol. 2, no. 8, Aug. 2010, pp. 609–24. [11] E. Dillon et al., “PERIMETER: A Quality of Experience Framework,” Int’l. Wksp. Future Internet of Things and Services, 1 Sept. 2009, Berlin, Germany. [12] Platform for Telecommunications Services Analysis (PASITO), http://www.hpcn.es/projects/pasito/. [13] A. Bianco et al., “Multistage Switching Architectures for Software Routers,” IEEE Network, vol. 21, no. 4, July 2007, pp. 15–21. [14] S. Figuerola et al., “PHOSPHORUS: A Single-Step OnDemand Services Across Multi-Domain Networks for eScience,” Network Architectures, Management, and Applications, Proc. SPIE, vol. 6784, 2007.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
[15] EC co-funded project GN3, Joint Research Activity 2, http://www.geant.net/Research/Pages/home.aspx.
BIOGRAPHIES P ETER S ZEGEDI (
[email protected]) ____________ received his M.Sc. degree in electrical Eengineering from Budapest University of Technology and Economics (Hungary, 2002). He then worked toward a Ph.D. in the Department of Telecommunications. His main research interests include design and analysis of dynamic optical networks, especially optical Ethernet architectures, network virtualization, control, and management processes. He worked for Magyar Telekom (2003–2007) and then joined TERENA in January 2008. MARCO RUFFINI (
[email protected]) ___________ is currently a lecturer on optical network architectures at the Department of Computer Science of the University of Dublin, Trinity College, where he obtained his Ph.D. in 2007. He is part of the CTVR telecommunication research centre, and his research interests include low-based optical switching, experimental optical testbeds, and techno-economic studies of next-generation transparent architectures. He worked for Philips Research Laboratories in Aachen (2003–2005), and before that as a technology consultant for Accenture, Milan. He holds a degree in electronic engineering from Universitá Politecnica delle Marche in Italy (2002). D ONAL O’M AHONY (
[email protected]) _______________ graduated with first class honors in engineering from Trinity College in 1982. After a brief career in industry at Sord Computer Systems in Tokyo and IBM in Dublin, he rejoined Trinity College as a lecturer in computer science in 1984, completing his Ph.D. in the area of software reusability in 1990. At Trinity, he built up a successful research group in networks and telecommunications. He spent the year of 1999 as a Fulbright Fellow at Stanford University, California, before returning to his present position as professor in computer science at Trinity College. In July 2004 he led a team to establish CTVR, a major multi-university research centre established in association with Bell Labs. JORDI FERRER RIERA (
[email protected]), ____________ M.Sc., graduated in computer science from the Technical University of Catalonia (FIB-UPC). He joined the i2CAT foundation as a software engineer in early 2007 for developing the MANTICORE project. In 2008 he started his collaboration in the PHOSPHORUS project. He also collaborates with the GLIF Generic Network Interface Technical Group (GNI-TG), adapting the Harmony Service Interface (HSI) to the GNI specification. In 2010 he started collaborating actively in the GEYSERS project. He is currently a Ph.D. candidate in the Telematics Engineering Department of UPC. J OAN A NTONI G ARCIA -E SPIN (
[email protected]) _________________ holds an M.Sc. from UPC (2007). He is a research project manager at the Network Technologies Cluster of the i2CAT Foundation and Ph.D. candidate at UPC. He wrote his Master’s thesis on the design and implementation of TE-enabled, DiffServ-aware MPLS networks for providing end-to-end QoS. He is currently working in European projects such as GÉANT3 and GEYSERS and participated in PHOSPHORUS and FEDERICA, as well as several Spanish research projects in the past. He is an active contributor to and editor of the Network Service Interface standard in the Open Grid Forum. MARKUS HIDELL (
[email protected]) _________ is an assistant professor in communication systems at the Royal Institute of Technology (KTH), Sweden. He received his M.Sc. degree in telecommunication systems and his Ph.D. degree in telecommunication from KTH in 1992 and 2006, respectively. Since January 2008, Markus is an assistant professor at Telecommunication Systems Laboratory (TSLab), KTH. His current research interests include switch and router architectures, protocols, and network architectures. He has coauthored six patents in the area of network resource management and network topologies. PETER SJÖDIN (
[email protected]) ______ holds a Ph.D. from Uppsala University in computer science. Since December 2002 he is an associate professor in communications networks at KTH in TSLa) of the School of Information and Communication Technology. His current research interests include network systems, protocols and network architectures, virtualization architectures, and residential service management. He holds five patents in the areas of fast lookups and network resource management, and network topologies.
PEHR SÖDERMAN (
[email protected]) _______ received his M.Sc. in engineering from KTH in 2008, specializing in computer science. He joined the TSLab research group at KTH in 2009, where he is currently a Ph.D. student. His current research interests include measurement methodology, experiment repeatability, and security in large-scale research networks.
IEEE
BEMaGS
______________ is an associate A NDREA B IANCO (
[email protected]) professor in the Electronics Department of Politecnico di Torino, Italy. He was Technical Program Co-Chair of HPSR 2003 and 2008, DRCN 2005, and the IEEE ICC 2010 ONS Symposium. He was a TPC member of several conferences, including IEEE INFOCOM, IEEE GLOBECOM, and IEEE ICC. He is editor of the Elsevier Computer Communications journal. His current main research interests are in the fields of protocols and architectures for all-optical networks and switch architectures for high-speed networks.
L UCA G IRAUDO (
[email protected]) _____________ graduated from Politecnico di Torino with a degree in telematics engineering in April 2007. Between June and December 2007 he was involved in the design and implementation of distributed software router architecture (BORA-BORA project). Since January 2008 he has been with Politecnico di Torino as a Ph.D. student under the supervision of Prof. Andrea Bianco, working on software routers, network virtualization, resource distribution, and centralized control in networks.
F
The European NREN partners of the former project consortium have agreed to continuously support the operation of the infrastructure and are willing to do further development under the recently launched “Network Factory” task of the GN3 project’s Joint Research Activity 2.
CRISTINA CERVELLO-PASTOR (
[email protected]) ____________ received her M.Sc. degree in telecom engineering and Ph.D. degree in telecommunication engineering, both from the Escola Tècnica Superior d’Enginyers de Telecomunicació, UPC, Barcelona, Spain. She is currently an associate professor in the Department of Telematics Engineering at UPC, which she joined in 1989, and leader of the optical networks research group within the BAMPLA group. V ICTOR L OPEZ (
[email protected]) ____________ received his M.Sc. degree in telecommunications engineering from Universidad de Alcalá de Henares (2005) and his Ph.D. degree in computer science and telecommunications engineering from Universidad Autónoma de Madrid (UAM) in 2009. In 2004 he joined Telefónica I+D, where he was involved in next-generation networks for metro, core, and access. In 2006 he joined the High-Performance Computing and Networking Research Group at UAM. Currently, he is an assistant professor at UAM, where he is involved in optical metro-core projects (BONE, MAINS). His research interests include Internet services integration over optical networks (OBS solutions and multilayer architectures). MIGUEL PONCE DE LEON (
[email protected]) ___________ is a Waterford Institute of Technology (WIT) graduate with a degree in electronic engineering, and he is currently head of Communication Infrastructure Management, a unit of the TSSG, a research group that focuses on telecommunications software services management and Internet technologies. He has participated in over 35 international re-search projects focusing on the future Internet, where his research team of 30 staff concentrate on the architecture and design of the future Internet, looking specifically at self-management of virtual resources, threats to communication-based service information, and the Living Labs concept. He was General Chair for TridentCom 2008 and is a member of FITCE. GEMMA POWER (
[email protected]) _________ holds a first class honors degree from WIT and has extensive industry experience with Apple. Since joining the TSSG in 2007, she has worked on the EU MORE project in the area of wireless sensor networks and validation and test. She is currently working on the EU PERIMETER project where her primary focus is the testbed setup, configuration, and federation, in addition to the code integration and testing, proof of concept, and demonstration aspects of the project. SUSANNE NAEGELE-JACKSON (susanne.naegele-jackson@rrze. _________________ uni-erlangen.de) ________ graduated with an M.Sc. in computer science from Western Kentucky University and the University of Ulm, Germany. She received her Ph.D. (Dr.-Ing.) from the University of Erlangen-Nuremberg in computer science. She has worked at the Regional Computing Center of the University of Erlangen-Nuremberg since 1998 on a variety of national and international research projects such as GTB, Uni-TV, Uni-TV2, VIOLA, MUPBED, EGEE-III, GN3, FEDERICA, and NOVI. She has authored and co-authored over 30 scientific publications, and teaches classes on multimedia networking at the Regional Computing Center.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
61
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
FUTURE INTERNET ARCHITECTURES: DESIGN AND DEPLOYMENT PERSPECTIVES
Content, Connectivity, and Cloud: Ingredients for the Network of the Future Bengt Ahlgren, Swedish Institute of Computer Science Pedro A. Aranda, Telefónica, Investigación y Desarrollo Prosper Chemouil and Sara Oueslati, Orange Labs Luis M. Correia, IST/IT-Technical University of Lisbon Holger Karl, University of Paderborn Michael Söllner, Bell Labs/Alcatel-Lucent Annikki Welin, Ericsson Research
ABSTRACT A new network architecture for the Internet needs ingredients from three approaches: information-centric networking, cloud computing integrated with networking, and open connectivity. Information-centric networking considers pieces of information as first-class entities of a networking architecture, rather than only indirectly identifying and manipulating them via a node hosting that information; this way, information becomes independent from the devices they are stored in, enabling efficient and application-independent information caching in the network. Cloud networking offers a combination and integration of cloud computing and virtual networking. It is a solution that distributes the benefits of cloud computing more deeply into the network, and provides a tighter integration of virtualization features at computing and networking levels. To support these concepts, open connectivity services need to provide advanced transport and networking mechanisms, making use of network and path diversity (even leveraging direct optical paths) and encoding techniques, and dealing with ubiquitous mobility of user, content and information objects in a unified way.
INTRODUCTION The Internet’s architectural model has sustained continuous development for the past four decades and provided an excellent substrate for a wide range of applications. The amount of mobile data has been growing exponentially, and one can expect this tremendous growth to con-
62
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
tinue. Despite the Internet’s uncontested successes, some challenges for this model are becoming apparent, like adding applications more complex than simple client/server or peerto-peer ones (e.g., multitier), or deploying information-centric ones distributed over different providers; moreover, the range of so-far successful business models seems limited. Also, coordinating and integrating more diverse technologies, networks, and edge devices is getting overly expensive, and security issues are becoming real barriers to deployment and use. Information itself has become more and more important in all aspects of communication and networking. Most of the traffic in today’s Internet is related to content distribution, which includes file sharing, collaboration applications, and media streaming, among others. The interaction patterns of emerging applications no longer involve simply exchanging data end-toend. These new patterns are centered on pieces of information, being accessed in a variety of ways. Instead of accessing and manipulating information only via an indirection of servers hosting them, putting named information objects themselves at the center of networking is appealing, from the viewpoint of information flow and storage. This information-centric usage of the Internet raises various architectural challenges, many of them not being handled effectively by the current network architecture, which makes information-centric networking an important research field. In this new paradigm, storage for caching information is part of the basic network infrastructure, a network service being defined in terms of named information objects (e.g., web pages, photos, movies, or text documents), inde-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
pendently of where and how they are stored or transported. This approach is believed to enable an efficient and application-independent largescale information distribution. Another problem is related to network applications, which can fluctuate rapidly in popularity and in terms of the amount of user interaction. This makes provisioning of both server and storage, as well as of networks, a difficult problem. On the server and storage side, cloud computing has successfully addressed many of these challenges, using virtualization as a core technique. However, it is still unclear how to provide suitable network support for such highly variable applications when they run not just over the tightly controlled, custom-tailored network of a cloud computing operator, but rather inside more complex and diverse operator networks. In such a network, it might be possible to provide the computational resources, but it is not obvious how to dynamically provide the necessary networking support/capacity or the complex networking topology required. Furthermore, security in both networks and cloud computing is a key challenge to success. One needs an integration of network resource management with cloud computing, an integration of provisioning distributed cloud resources with the network services to connect such distributed resources reliably at a required quality. This combination is called cloud networking. Transport of information is another matter that needs to be addressed. In the current Internet, transport relies on connectionless forwarding of small data packets that is not able to exploit the additional (semantic) information that is available in the end or edge systems; additionally, it is incapable of making use of the context information that defines and controls related flows throughout different network aggregation layers, leveraging the capabilities of heterogeneous transmission technologies. For example, it is practically impossible to exploit the diversity existing over different communication technologies between two endpoints (e.g., random variations in channel quality or structural differences in channel properties, like different delay/data rate trade-offs), switching between technologies as the flow’s required data rate changes. Similarly, efficient multi-path/protocol/layer optimization is still unfeasible. In order to efficiently use such high-speed future network technologies, it is critical to implement cross-layer coordination with new interdomain transport, switching, and routing protocols. Furthermore, the current Internet is a flat, serviceneutral infrastructure; this is reflected in today’s rigid peering agreements, which limit the type of business models and service agreements that can be applied at interprovider interfaces. In today’s cellular networks, the introduction of new services is a cumbersome process due to the complexity of setting up the necessary roaming agreements, since different networks may have different releases and features (besides the billing problem). Open connectivity offers an approach to address these problems. The aspects addressed above can be put into a perspective of joint planes for a new architecture (Fig. 1). Three approaches are addressed in
Cloud networking aspect
IEEE
BEMaGS F
Information-centric networking aspect
Open connectivity aspect
Figure 1. Three aspects of a new network architecture.
the current article: information-centric networking, cloud networking, and open connectivity. The next sections present these concepts and discuss them in detail.
A TARGETED SCENARIO A scenario can help to put some of the previously mentioned aspects into perspective, and explicitly show an integrated approach to the problem. Obviously, a single example scenario cannot convey all aspects of a complex situation, but it is important to understand how the various approaches can fit together. Consider a user, Alice, offering some piece of information (be it static or dynamic, e.g., streaming video) from her mobile handset to her content repository in the network. She shares this content, which becomes unusually popular, being viewed by many people, Alice’s “followers,” most of whom use different network operators, thus causing a large amount of relatively slow and quite expensive cross-operator traffic (Fig. 2a). This situation creates some incentive for the network operator to improve this content delivery situation (out of self-interest, but also to improve user-perceived quality). A possible solution is the usage of the network-centric architecture, together with some open connectivity services, so that the increased load causes additional instances of Alice’s content repository to be quickly spun up within some of these other operators’ own networks (Fig. 2b). The replication of the popular information to another location is facilitated by information-centric caching mechanisms. If necessary, this infrastructure, with the necessary processing means, ensures that information is processed at and delivered from topologically advantageous places — unlike today’s cloud computing, where the processing can take place far away, with long round-trip delays. This allows for reduction in
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
63
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Alice’s operator
Another operator
Alice’s operator Another operator Cloud networking infrastructure Request
Alice
F
Alice’s operator Another operator Cloud networking infrastructure
Alice Video
Video
BEMaGS
Create
Alice Video
A
Video Video
Video
“Followers”
“Followers”
(a)
(b)
“Followers” (c)
More “Followers”
Figure 2. Three steps in an advanced user-content-provisioning scenario. cross-operator traffic, since each of Alice’s followers can now access information using only network operator local traffic, and Alice’s video is only replicated once between her operator and each of the other operators. However, this opens some transport problems: by the time the additional nodes are operational, a substantial amount of video may already have been buffered at Alice’s (home network) node, which can cause problematic delays for new followers; existing followers that have been receiving cross-operator traffic will need to switch to their now-local instance of Alice’s node. These problems may be addressed by multipath transport connectivity, which can handle the transport of the initial (previously buffered) video burst via higher-bandwidth links for interoperator traffic before seamlessly falling back to the cheaper connectivity that is sufficient to keep up with Alice’s ongoing video stream (Fig. 2c). Hence, storing, processing, and transporting information turns into an integrated problem, while today only isolated solutions are available.
INFORMATION-CENTRIC NETWORKING THE NOTION The notion of information-centric networking (ICN) has been proposed by several initiatives in the last few years. The core idea of most proposals is to consider pieces of information as the main entities of a networking architecture, rather than only indirectly identifying and manipulating them via a node hosting that information. Thus, information becomes independent of the devices in which it is stored, enabling efficient and application-independent information caching in the network. This approach is believed to result in a network that is better adapted to information distribution and retrieval, which are the prevailing uses of current network technologies. Notable examples are the work on contentcentric networking (CCN) [1], publish/subscribe schemes [2], directly embedding publish/subscribe schemes into the network fabric (PSIRP project) [3], the NetInf work by the 4WARD project [4], upon which our own ongoing work is mostly based, or, earlier, the DONA project [5]. Similar ideas have also been considered in the context of wireless sensor networks (e.g., the idea to use predicate-based “interests” to identify which data
64
Communications IEEE
shall be transported, with protocols like directed diffusion [6] realizing that idea).
AN EXAMPLE ARCHITECTURE: NETINF Let us consider one of the approaches in more detail. The NetInf ICN architecture developed in the 4WARD project comprises three major components: a naming scheme for information objects (IOs), a name resolution and routing system, and in-network storage for caching. These components are illustrated at a high level in Fig. 3 and described in the following paragraphs. The naming scheme is important for making information objects independent of the devices storing them. The hard part is not to make the names (information object identifier, IO ID in the figure) location-independent, but rather to fulfill the security requirements that result from the location independence. One cannot depend on host-based authentication of a delivering server, since one wants any node in the network, dedicated caches as well as end hosts, holding a copy to be able to share that with others. In order to be able to trust a copy of an information object coming from an untrusted device, the receiver must be able to independently verify the integrity of the object so that it becomes impossible to make forgeries. Therefore, the naming scheme has a cryptographic binding between the name itself (using field A of the ID) and the object, similar to DONA. Furthermore, the naming scheme supports dynamic objects, owner authentication and identification, changing the owner of an object, and anonymous owners. The purpose of the name resolution and routing system is to link the object names (IO IDs) to the actual information objects, so they can be queried and retrieved. It is a major challenge for all information-centric approaches to design this resolution system so that it scales to the global level. The system can be viewed as a variant of the Internet’s Domain Name System (DNS). The system can also be viewed as a variant of IP routing, where query packets are routed toward the location of the resolution records, or all the way to an actual copy of the information object. NetInf supports both of these models by allowing different name resolution protocols in different parts of the network. Protocols based on distributed hash tables (DHTs) have been investigated as one suitable technology; multicastbased protocols have been investigated for implementing resolution in a local scope.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Storage for caching the actual bits of the information objects — the bit-level objects (BOs) in Fig. 3 — is an integral part of the network service. Potentially, BOs are cached at all routers and end systems in the network. The goal is to deliver the requested object from the best cache(s) holding a copy to the client. A cached copy can be found either through the name resolution system or by a cache-aware transport protocol. The NetInf application programming interface (API) is inspired by publish/subscribe. A producer of information can publish an information object, creating a binding in the name resolution system, and revoke a publication, removing the binding. A consumer can resolve an information object name, returning the corresponding binding(s), and retrieve an object using the information in the binding(s).
IO ID: Type A=hash(PKIO)
A
BEMaGS F
L=label
Name resolution system
Storage/caching
Name resolution records
Name resolution
Type Meta data
Metadata
Content
IO
IO ID’
BO
LocatorA
A BO
B BO
BO
LocatorB
Figure 3. Major components of the 4WARD NetInf architecture.
COMPARISON OF APPROACHES Other approaches make different design choices. Table 1 summarizes the main conceptual differences for four exemplarily chosen popular ICN variants, previously mentioned. The main aspects of difference are: • The choice of what to consider a piece of information with the corresponding naming model • Whether and how names are resolved into routable addresses of a simpler system (like IP) or whether name-based routing is used • How transport and caching are integrated
CHALLENGES IN ICN The ICN approach is still young, with many remaining research challenges, some of the most important ones being outlined in what follows. Global scalability: An ICN needs to handle on the order of 1015 unique information objects at the global scale. Some solutions have been proposed (e.g., using DHTs), and calculations have been made suggesting that it is feasible to construct a global name resolution/routing system meeting this requirement. It still remains to be proven by experiments using real implementations. Cache management: Resource management needs to go beyond considering link capacity, and has to address, in particular, cache storage. Some control of caching is needed to deliver a predictable service. Different cache replacement algorithms might be needed for different applications and usage patterns. Cache management protocols are needed for, say, collaboration between caches. Performance models are needed, accounting for distributed caching, statistical features of queried pieces of information (popularity, content size, usage patterns, correlations between objects), and the interplay between caching and data rate, notably for dimensioning. Congestion control: ICNs depart from today’s Internet in two ways: they are receiver-oriented, and they change the end-to-end principle. While the former implies that end users may control the rate of information delivery, the latter creates an opportunity for implementing congestion control protocols between relevant nodes inside the network, through (chunk) query message pacing. This pacing mechanism may, for exam-
ple, be enforced between border routers of two different network providers in a consistent manner with respect to the charging model in use for information transport. Deployment issues: To deploy informationcentric schemes, there must be both incentives for users and operators, as well as the technical feasibility to introduce them. For operators, the appeal might lie in new business models (act as information host, cache provider) and operational advantages (reduce interoperator traffic, since information has to be exchanged only once between two operators). Incremental deployment is also a sine qua non condition; it is facilitated by schemes that can use existing routing and forwarding infrastructures (e.g., like NetInf can use different name resolution systems as plug-ins and directly run on top of IP as well as on lower layers).
CLOUD NETWORKING CLOUDS ARE RESTRICTIVE Provisioning of data processing and storage in clouds sitting at the edge of a network has proven to be extremely useful for a wide range of conventional applications; it is also a model that is well in accordance with today’s network architecture. But when one considers either more demanding applications (e.g., with stringent latency requirements) or an advanced networking architecture, there are reasons to rethink the current cloud model. Consider ICN as a case study: ICN requires storage and computing facilities distributed into the network at a very fine granularity level, in particular, if it has to go beyond pure content distribution services and embrace active information objects. Leveraging current cloud computing solutions, based on server farms sitting at the edge of the network, provides an insufficient level of flexibility and performance, in particular, latency, to the end user. Serving ICN requests from the edge of the network will not result in acceptable performance, hence, while ICN will require cloud-like functionality, the notion of cloud computing has to be reconsidered. This implies the need to embed computation and
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
65
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Design aspect
CCN
NetInf
PSIRP
DONA
Naming and security of information objects
Hierarchical, need to trust signing key to establish integrity
Flat, self-certifying; support for versioning, and transfer of ownership
Flat, self-certifying; notion of scope
Flat, self-certifying
Name resolution and routing
Name-based routing using longest prefix of hierarchical names
Allows both resolution using, e.g., DHTs, and name-based routing
Name resolution using a rendezvous function, within a specified scope
REGISTER and FIND primitives; hierarchical resolution handlers
Transport and caching
Transport using namebased routing; finds cached objects through local search as well as on the path to the publisher
Allows multiple transport protocols; finds cached objects through name resolution as well as cache-aware transport
Transport routing and forwarding, using separate forwarding identifiers
Caching in resolution handlers
A
BEMaGS F
Table 1. Comparison of different concepts.
storage deeply into the network to provide the required quality of experience. A cloud system serving an ICN architecture has to create ICN instances at various places in the network (and not just outside), and it has to provide these instances with a suitable and secure, possibly private, network. Hence, one needs to integrate cloud and (possibly virtual) networking services into cloud networking. More generally, in a traditional cloud, massive amounts of data will be “sitting in the cloud,” waiting to be accessed by users anywhere and anytime. “Sitting in the cloud” also implies the need for a higher level of flexibility in the network: on the one hand, applications will reside in it, will be massively distributed (even over several cloud centers), and will be accessible to a massive number of users; on the other, the network itself will be composed of a vast range of different network infrastructures, which will be undoubtedly managed by different operators. These requirements are not new, and the TeleManagement Forum had been addressing them for some time in initiatives like IPSphere. But today’s cloud solutions are based on concepts inherited from grid computing, and as such, they do foresee massive deployment of computing resources located at the edge of the network in general. Advanced solutions for distributed services were inspired by the grid (e.g., Eucalyptus), but will not serve our purpose either. They implement the Infrastructure as a Service (IaaS) paradigm, being massively based on pushing computing and content to virtual machines localized at a few locations at the edge of the network, i.e., at large data centers. None of these approaches is suitable to act as an execution platform for ICN, where both storage and computing will be distributed, yet might still heavily interact with each other. For a pervasive deployment of the kind of infrastructure one is aiming at, there is the need to provide the network with mechanisms to access and to create such computing resources at any place in the network they might be deployed at. More important, tighter integration with virtualization at all possible levels is necessary: applications in the cloud run in parallel, sharing the infrastructure, and need to be isolated from one another to provide predictable security and
66
Communications IEEE
performance guarantees at all levels, including the network plane. Additionally, current approaches to network virtualization are too static: virtual private networks (VPNs) at layers 2 and 3 are conceived as semi-static entities, which require often manual intervention when connections to end-user locations are created or destroyed. Signaling, a possibility to make the operation of VPNs more dynamic, is currently based on protocols like BGP-4, which have been designed to minimize the oscillation probability in the infrastructure, and therefore are not too dynamic.
BENEFITS OF AN INTEGRATED APPROACH One needs a solution that distributes the cloud (and its main benefits, on-demand availability of computing and storage with massive benefits of scale) more deeply into the network and disperses the cloud closer to the end user to reduce latency: one might talk about mist computing instead of cloud computing (Fig. 4). Moreover, these “misty” resources need to be flexibly networked across a backbone network, with isolation and security in place, the allocation of storage, computation, and networking connectivity between them becoming an integrated problem — applications can only be mapped onto a part of a cloud, when the required networking resources are in place, both to other parts of a cloud and to the end-user population the cloud part is intended to serve. The approach proposed here intrinsically takes the finer level of granularity needed in order to implement an infrastructure into account, which is highly responsive and provides a tighter integration of virtualization features at computing and networking levels, possibly trading off computing and networking against each other (e.g., use slower computing nearby vs. fast computing far away). The levels of envisioned adaptability provide better adaptation to network conditions and higher robustness to flash crowd effects. The network will adapt the amount of resources to the traffic needs, and move computational resources nearer to the physical locations where they are needed, creating the required network connectivity on demand. Additionally, it will support a rich ecosystem of middleware, which can
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
F
Some lines of current research in cloud Cloud 1
Cloud 1
networking focus on optimizing networking inside big cloud
Internet
Internet
data centers, since measurements in such environments show that most of
Cloud 2 (a)
(b)
the traffic stays within the data
Figure 4. From cloud (a) to mist (b) computing, supported by cloud networking: resources of cloud 2 (shown in green) are spread much finer and deeper into the network, close to the actual point of usage.
centre. The envisioned architecture does not geographi-
be run concurrently and isolated in different virtual infrastructures; cloud networking is not meant as an exclusive runtime environment for ICN alone, but as a generic service accessible for many different kinds of applications that need to run in the network with similar levels of adaptability and scale. For example, the Software as a Service (SaaS) paradigm should also benefit from cloud networking, moving the provided software away from today’s centralized and remote data centers closer to the customer. Moving closer to the customer is in the interest of resource efficiency too: usage dictates the network portions that are activated for a specific service.
CHALLENGES Some lines of current research in cloud networking focus on optimizing networking inside big cloud data centers [7], since measurements in such environments show that most of the traffic stays within the data center [8]. The envisioned architecture does not geographically confine traffic in this way. The impact of traffic patterns associated with cloud networking applications [8] needs to be studied in highly distributed scenarios as considered here. Another challenge that arises in massively distributed environments is failure protection: an application at a given network location might not work as expected. This situation needs to be detected and corrected. Approaches like the one presented in [9] need to be explored.
OPEN CONNECTIVITY CHALLENGES OF INTERNET TRANSPORT AND CONNECTIVITY ARCHITECTURES So far, the current Internet transport paradigm focused to a large extent on the provisioning of a transparent TCP/IP based point-to-point connectivity between addressable hosts irrespective of the underlying transport technologies. However, there is a tremendous increase in capacity in the lower-level network technologies (fiber, copper, and wireless technologies), but the usable network capacity increasingly lags behind the demands of emerging resource-hungry net-
worked applications (created by, e.g., content distribution, cloud computing, or social networking). In addition, the heterogeneity of deployed network technologies makes it hard to exploit the particular network resources and features on an end-to-end, or even edge-to-edge, basis for the sake of new evolutions, like ICN or cloud networking. Therefore, there is also a need for an advanced open connectivity service framework that addresses the issues in the transport mechanisms of a Future Internet. It aims at leveraging advanced features (e.g., multipoint, multipath, dynamic switching, and extended bandwidth) of link technologies, especially of optical transport networks, making use of network (and path) diversity and advanced encoding techniques, and at dealing with ubiquitous mobility of user, content and information objects in a unified way. Access to these mechanisms should be provided through new open and extensible interfaces, between client (user) and network, as well as between networks. Figure 5 presents multilayer transport architecture and interfaces for open connectivity services. While the physical networks offer an evergrowing optical bandwidth, and tend to aggregate links and switching/routing capabilities as much as possible for efficiency purposes in the core network, connectivity for the ICN approach will require high-performance distribution, referencing and managing a large number of interlinked, but relatively small, information chunks located all over the world, preferably at the edge of the network. Today, this seems like diverging interests, a challenge that needs to be addressed in an evolved transport system. Therefore, the use cases for open connectivity will include the special needs for Wide Area Networks interconnectivity of new players, like distributed service centers and large enterprises (acting as information-centric nodes or cloud service providers), providing them with advanced and easy-to-use open APIs, to set up and efficiently control their private “virtual cloud networks” across multiple transport technologies and domains. Data centers in such a context will also comprise “mobile operation centers,” such as traditional IP Multimedia Subsystem (IMS)
cally confine traffic in this way.
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
67
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Data center
Cloud networking, information-centric networking Virtualized node
Service router
API
API
Optically switched networking (OTN) Packet/ label router
Sub-λ switch
Control and management
Open UNI
Control and management
Open NNI
Open UNI
“Electrical” networking (IP/MPLS)
All-optical networking
λ switch
Fiber switch
λ switch
Figure 5. Open Connectivity Services: multi-layer transport architecture and interfaces (UNI: user-to-network interface, NNI: network-tonetwork interface). network functionalities running in a specific “mobility cloud.”
LEVERAGING LOWER-LAYER TRANSPORT EFFICIENCY The current Internet is not capable of making use of context information that defines and controls related flows throughout different network aggregation layers, leveraging the capabilities of heterogeneous transmission technologies, including IP/multiprotocol label switching (MPLS), WiFi, third generation/Long Term Evolution (3G/LTE), Ethernet, and optical networks. For example, TCP end-to-end transport with error protection, flow control, and congestion control is completely decoupled from the routing and forwarding aspects of interconnected networks. This architecture does also not allow leveraging advanced features of upcoming global network technologies, such as carrier-grade Ethernet or advanced optical switching techniques (e.g., concerning path management, resilience, or quality of service [QoS] mechanisms). For efficiently utilizing such high-speed future network technologies, it is critical that there is cross-layer coordination with new interdomain transport, switching, and routing protocols [10]. The evolution of transport networks is mainly driven by advances in optical transmission technologies, increasing the usable transmission bandwidth in optical fibers, as well as by the evolution of photonic integrated circuit technologies and the electrical processing in silicon, more and more used for switching of sub-lambda, single wavelengths and wavebands in a dynamical way [11]. The ability of using direct lightpaths and optical technologies for traffic off-loading the Internet core, and reducing (electrical) processing in intermediate hops, will have a beneficial
68
Communications IEEE
impact on the energy budget of the global Internet overall, a problem being recognized only in recent years in the context of so-called green ICT [12]. However, this requires a new modified addressing and routing architecture, if a packet should be processed in less electronic steps, and then put into an optical path that ends up near the destination, in order not to run into routing table explosion and scaling problems caused by the extensive use of multilayer techniques across multiple domains [13].
RELATED WORK The design of a new transport architecture for the Future Internet has partly been addressed in previous projects. Most notably, the European project 4WARD (http://www.4ward-project.eu) has developed a functional architecture for generic paths that enables new forms of in-network processing. The 4WARD Generic Path concept encapsulates both interfaces and functionality central to data transport in an objectoriented fashion, thereby enabling access to routes, paths and transport functionalities, allowing to stack and connect them dynamically (e.g., to implement distributed mobility management). Such principles can now be extended to develop a lightweight, flow-aware concept of routing across and managing of rich communication paths (i.e., multipoint, multiprotocol, multipath). This should allow for the development of open connectivity services that satisfy the needs of flash network slices (i.e., capable of being virtualized) and ICN (i.e., connecting content and data centers). The European project Trilogy (http://www. trilogy-project.org) proposes a unified architecture for control, divided into a delivery service and transport services. The delivery service is composed of a reachability plane, responsible for the outgoing link selection, enabling network-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
wide reachability, and a resource plane, responsible for sharing the transmission resource between packets. The transport services provide functions for reliability, flow control and message framing. Likewise, alternative control plane architectures arise out of the ongoing future Internet research activities, such as OpenFlow (http://www.openflowswitch.org), which provides a new interface to Ethernet switches, enabling experiments with novel protocols at flow level on the one hand, and allows for programmable networking on the other. Further related activities exist in other ongoing European projects working towards the Future Internet, such as ETICS, GEYSERS and STRONGEST; generally, all these projects deal with variants of multi-layer network architectures and their multi-domain interconnection. To our knowledge, only the SAIL project (http://www.sail-project.eu) focuses on providing the means for user/application controlled access to establish connectivity tailored to the needs of future internet stakeholders, such as cloud networking or ICN.
CONNECTIVITY AS AN OPEN SERVICE The novel proposed open connectivity approach will extend the current point-to-point related connectivity toward multi-p* (i.e., multipath/ point/protocol) transport and routing, investigating the interactions between multi-p* transport, path selection and routing, and having an endto-end cross-layer and cross-domain approach for multi-p* management. The proposed solution is based on a multidomain architecture, allowing for open and extensible communication and cooperation between the control planes of different network domains (user-to-network, and network-to-network) in a unified way. It enables the generic exchange of resource information for data flows across technology boundaries, in order to support content and information delivery in ICN, and provide appropriate dynamic and virtualized connectivity for cloud networking. That will also allow endto-end optimization concerning transport energy efficiency or intelligent sharing of network resources, caches and data processing. As an application, one expects the rise of specific cloud services provided by distributed and interconnected data centers, such as mobile access to community and social networks running in a specific “mobility cloud.” Networking of such mobility centers might require new forms of mobility management that go beyond serving mobile users in the access networks, and include mobility of content and information within the interconnected distributed operation centers. A significant efficiency benefit can be expected by making use of path diversity, in both advanced optical and wireless network technologies. A promising alternative to state-of-the-art mobility management, with its single centralized anchor point, is a dynamic distributed mobility management [14]. In the cloud, the network dynamically chooses the optimal location of mobility service anchor points on a per-user/per-device or even per-flow basis. Open connectivity services will enable the cloud to make use and manage the
multi-flow and multi-path routing capabilities provided edge-to-edge across the networks.
IEEE
BEMaGS
CONCLUSIONS The Internet has been based up to now on an architectural model that has coped with a sustained continuous development and provided a good environment for a wide range of applications. Nevertheless, challenges for this model became apparent, namely at the applications level, not only from the technical viewpoint but also from the business one. This article addresses aspects of a new architecture, from three approaches: information-centric networking, cloud networking, and open connectivity services. Information-centric networking considers pieces of information as main entities of a networking architecture, rather than only indirectly identifying and manipulating them via a node hosting that information; this way, information becomes independent from the devices they are stored in, enabling efficient and applicationindependent information caching in the network. Major challenges include global scalability, cache management, congestion control, and deployment issues. Cloud networking offers a combination and integration of cloud computing and virtual networking. It is a solution that distributes the benefits of cloud computing more deeply into the network, and provides a tighter integration of virtualization features at computing and networking levels. Current challenges encompass the optimization of networking inside cloud data centers, the study of the impact of traffic patterns associated with cloud networking applications in highly distributed scenarios, and failure protection in massively distributed environments. Open connectivity services address transport mechanisms issues, aiming at leveraging advanced features of link technologies, namely in optical networks, making use of network (and path) diversity and advanced encoding techniques, and at dealing with ubiquitous mobility of user, content and information objects in a unified way. Challenges address, among others, the development of a lightweight flow-aware concept of routing across and managing of multipoint/protocol/path communications, satisfying the needs of flash network slices, supporting content and information delivery in informationcentric networks, and providing appropriate dynamic and virtualized connectivity for cloud networking.
F
Cloud networking offers a combination and integration of cloud computing and virtual networking. It is a solution that distributes the benefits of cloud computing more deeply into the network, and provides tighter integration of virtualization features at the computing and networking levels.
ACKNOWLEDGMENTS The authors would like to thank all their colleagues from the SAIL project team. Without their contributions, this article and the insights behind it would not have happened. This work has been partially funded by the European Commission, under grant FP7-ICT-2009-5-257448SAIL.
REFERENCES [1] V. Jacobson et al., “Networking Named Content,” Proc. CoNEXT’09 — 5th Int’l. Conf. Emerging Networking Experiments and Technologies, Rome, Italy, Dec. 2009.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
69
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
[2] P. T. Eugster et al., “The Many Faces of Publish/Subscribe,” ACM Computing Surveys, vol. 35, no. 2, June 2003, pp. 114-31. [3] A. Zahemszky et al., “Exploring the Pub/Sub Routing & Forwarding Space,” Proc. IEEE ICC Wksp. Networks of the Future, Dresden, Germany, June 2009. [4] B. Tarnauca and S. Nechifor, Eds., Netinf Evaluation, EC FP7-ICT-4WARD Project, Deliv. D-6.3, June 2010 (http://www.4ward-project.eu). [5] T. Koponen et al., “A Data-Oriented (and Beyond) Network Architecture,” Proc. ACM SIGCOMM ’07, Kyoto, Japan, Aug. 2007. [6] C. Intanagonwiwat et al., “Directed Diffusion for Wireless Sensor Networking,” IEEE/ACM Trans. Net., vol. 11, no. 1, Feb. 2003, pp. 2–16. [7] M. Al-Fares, A. Loukissas and A. Vahdat, “A Scalable, Commodity Data Center Network Architecture,” Proc. ACM SIGCOMM ’08, Seattle, WA, Aug. 2008. [8] Arista Networks, Switching Architectures for Cloud Network Designs, http://www.aristanetworks.com/media/ system/pdf/SwitchingArchitecture_wp.pdf, _______________________ Apr. 2010, Architecting Low Latency Cloud Networks, http://www. ______________________________ aristanetworks.com/media/system/pdf/CloudNetworkLatency.pdf, _____ May 2009, Menlo Park, CA. [9] A. Carzaniga, A. Gorla, and M. Pezze, “Healing Web Applications Through Automatic Workarounds,” Int’l. J. Software Tools for Tech. Transfer, vol. 10, no. 6, Oct. 2008, pp. 493–502. [10] K. Sato and H. Hasegawa, “Optical Networking Technologies That Will Create Future Bandwidth-Abundant Networks,” IEEE/OSA J. Opt. Commun. and Net., vol. 1, no. 2, July 2009, pp. A81–A93. [11] D.T. Neilson, “Photonics for Switching and Routing,” IEEE J. Selected Topics in Quantum Electronics, vol. 12, no. 4, July–Aug. 2006, pp. 669–78. [12] J. Baliga et al., “Green Cloud Computing: Balancing Energy in Processing, Storage and Transport,” Proc. IEEE, vol. 99, no. 1, Jan. 2011, pp. 149–67. [13] G. J. Eilenberger et al., “Energy-Efficient Transport for the Future Internet,” Bell Labs Tech. J., vol. 15, issue 2, Sept. 2010, pp. 147–67. [14] F. Bertin, Ed., Description of Generic Path Mechanism based on Resource Sharing and Mobility Management, EC FP7-ICT-4WARD Project, Deliv. D-5.2.1, Dec. 2009 (http://www.4ward-project.eu).
BIOGRAPHIES B ENGT A HLGREN received his Ph.D. in computer systems in 1998 from Uppsala University, Sweden. He conducts research in the area of computer networking including the protocols and mechanisms of the Internet infrastructure. His main interest is the evolution of the Internet architecture, especially issues with naming and addressing on a global scale. Lately his research focus is on designing networks based on an information-centric paradigm.
70
Communications IEEE
A
BEMaGS F
P EDRO A. A RANDA obtained his Telecommunications Engineer title at Polytechnic University of Madrid’s (UPM’s) Telecommunications School, Spain. He joined Telefónica I+D in 1991 and is currently a technology specialist, conducting research in the areas of the future of the Internet and service agnostic networks. His main research interests are the design of Internet grade architectures and the behavior of BGP-4. Lately he has been working on the evolution of the Internet, especially issues related to interprovider and interdomain relationships. PROSPER CHEMOUIL [F’03] received his Ph.D. in control theory in 1978 from Nantes University. In 1980 he joined Orange Labs (then CNET), France Telecom’s R&D Centre, where he is currently director of a research program concerned with the design and management of future networks. LUIS M. CORREIA [SM’03] received his Ph.D. in electrical and computer engineering from IST-TUL in 1991, where he is currently a professor in telecommunications, with his work focused on wireless/mobile communications. He has been active in various projects within European frameworks. He was part of the COST Domain Committee on ICT and has been involved in Net!Works activities. HOLGER KARL received his Ph.D. in 1999 from Humboldt University Berlin; afterward he joined Technical University Berlin. Since 2004 he is a professor of computer networks at the University of Paderborn. He is also responsible for the Paderborn Centre for Parallel Computing, and has been involved in various European and national research projects. S ARA O UESLATI received her Ph.D. degree in computer science and networks from École Nationale Supérieur des Télécommunications, Paris, in 2000. She next joined France Telecom R&D as a research engineer in the field of performance evaluation and design of traffic controls for multiservice networks, and has led the Traffic and Resource Management research team since 2005. M ICHAEL S ÖLLNER is a technical manager at Alcatel-Lucent Bell Labs in Stuttgart, Germany. After he received a Ph.D. degree in applied mathematics (1984), he held various positions in the communication industry where he focused on systems engineering and research for network architectures and protocols, and currently for mobile systems beyond 3G and the future internet. He has been involved in various European cross-industry research projects. A NNIKKI W ELIN is a senior researcher at Ericsson Research, Packet Transport and Routing Department. She joined Ericsson in 1998. Her research interests include packet transport and overlay networks. She has co-authored more than 20 papers and over 20 patents. She has been active in various projects within European frameworks.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
FUTURE INTERNET ARCHITECTURES: DESIGN AND DEPLOYMENT PERSPECTIVES
PEARL: A Programmable Virtual Router Platform Gaogang Xie, Peng He, Hongtao Guan, Zhenyu Li, Yingke Xie, Layong Luo, Jianhua Zhang, and Yonggong Wang, Chinese Academy of Sciences Kavé Salamatian, University of Savoie
ABSTRACT Programmable routers supporting virtualization are a key building block for bridging the gap between new Internet protocols and their deployment in real operational networks. This article presents the design and implementation of PEARL, a programmable virtual router platform with relatively high performance. It offers high flexibility by allowing users to control the configuration of both hardware and software data paths. The platform makes use of fast lookup in hardware and software exceptions in commodity multicore CPUs to achieve highspeed packet processing. Multiple isolated packet streams and virtualization techniques ensure isolation among virtual router instances.
INTRODUCTION Deploying, experimenting, and testing new protocols and systems over the Internet have always been major issues. While, one could use simulation tools for the evaluation of a new system aimed at large-scale deployment, real experimentation in an experimental environment with realistic enough settings, such as real traffic workload and application mix, is mandatory. Therefore, easy programmable platforms that can support high-performance packet forwarding and enable parallel deployment and experiment of different research ideas are highly demanded. Moreover, we are moving toward a future Internet architecture that seems to be polymorphic rather than monolithic: the architecture will have to accommodate simultaneous coexistence of several architectures (the Named Data Network (NDN) [1], etc.) including the current Internet. Therefore, the future Internet should be based on platforms running different architectures in virtual slices enabling independent programming and configuration of the functions of each individual slice. Current commercial routers, the most important building blocks in the Internet, while attaining very high performance, offer only very limited access to researchers and developers for
IEEE Communications Magazine • July 2011
Communications IEEE
their internal components to implement and deploy innovative networking architectures. In contrast, open-software-based routers naturally facilitate access to and adaptation of almost all of their components, although frequently with low packet processing performance. As an example, recent OpenFlow switches provide flexibility by allowing programmers to configure the 10-tuple of flow table entries, enabling the change of packet processing of a flow. OpenFlow switches are not ready for nonIP-based packet flows, such as NDN. Moreover, while the switches allow a number of slices for different routing protocols through the FlowVisor, the slices are not isolated in terms of processing capacity, memory, and bandwidth. Motivated by these facts, we have designed and built a programmable virtual router platform, PEARL, that can guarantee high performance. The challenges are twofold: first to manage the flexibility vs. performance trade-off that translates into pushing functionality to hardware for performance vs. programming them in software for flexibility, and second to ensure isolation between virtual router instances in both hardware and software with low performance overhead. This article describes the PEARL router’s architecture, its key components, and the performance results. It shows that PEARL meets the design goals of flexibility, high performance, and isolation. In the next section, we describe the design goals of the PEARL platform. These goals can be taken as the main features of our designed platform. We then detail the design and implementation of the platform, including both hardware and software platforms. In the next section, we evaluate the performance of a PEARL-based router using the Spirent TestCenter by injecting into it both IPv4 and IPv6 traffic. Finally, we briefly summarize related work and conclude the article.
DESIGN GOALS In the past few years, several future Internet architectures and protocols at different layers have been proposed to cope with the challenges
0163-6804/11/$25.00 © 2011 IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
71
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Software platform VM 1
VM 2
VM 3
APPs
APPs
APPs
Protocol (IP/ Non-IP)
Protocol (IP/ Non-IP)
Protocol (IP/ Non-IP)
VM 5
VM 4
Net IO proxy
Generalized multicore CPUs
m
ea
am
m trea
st r
e str
IO stream
IO s
IO
IO
PCIE bus
VM N
DMA interface MAC and PHY
RX process TX process Hardware platform
Specialized packet processing card
Figure 1. Overview of the PEARL architecture. the current Internet faces [1]. Evaluating the reliability and performance of these proposed innovative mechanisms is mandatory before envisioning real-scale deployment. Besides theoretically evaluating the performance of a system and simulating these implementations, one needs to deploy them in a production network with a real user behavior, traffic workload, resource distribution, and applications mixture. However, a major principle in experimental deployment is the “no harm” principle that states that normal services on a production network should not be impacted by the deployment of a new service. Moreover, no unique architecture or protocol stack will be able to support all actual and future Internet services and we might need specific packet processing for given services. Obviously, a flexible router platform with high-speed packet processing ability and support of multiple parallel and virtualized independent architectures is extremely attractive for both Internet research and operation. Based on this observation one can define isolation, flexibility, and high performance as the needed characteristics and design goals of a router platform future Internet. In particular, the platform should be able to cope with various types of packets including IPv4, IPv6, and even non-IP, and be able to apply packet routing as well as circuit switching. Various software solutions like Quagga or XORP [2] have provided such flexible platform that is able to adapt their packet-processing components as well as to customize the functionalities of their data, control and management planes. However, these approaches fail to be fast enough to be used in operational context where a wire-speed is needed. Nevertheless, by adding and configuring convenient hardware packet processing resources such as FPGA, CPU cores and memory storage one can hope to meet the performance requirements. Indeed, flexibility and high performance are in conflict in most situations. Flexibility requires more functionalities to be implemented in software to maximize the programmability. On other hand, high performance cannot be reached in software and needs
72
Communications IEEE
A
BEMaGS F
custom hardware. A major challenge for PEARL is to allocate enough hardware and multi-cores in order to achieve both flexibility and high performance. Another design goal is relative to isolation. By isolation we mean a mechanism that enables different architectures or protocols running in parallel on separate virtual router instances without impacting each other’s performance. In order to achieve isolation, we should provide a mechanism to ensure that one instance can only use its allocated hardware (CPU cores and cycles, memory, resources, etc.) and software resources (lookup routing tables, packet queue, etc.), and is forbidden to access resources of other instances even when they are idle. We also need a dispatching component to ensure that IP or non-IP packets are delivered to specified instances following custom rules defined over medium access control (MAC) layer parameters, protocols, flow labels, or packet header fields. PEARL offers high flexibility through the custom configurations of both the hardware and software data paths. Multiple isolated packet streams and virtualization techniques enable isolation among virtual router instances, while the fast lookup hardware provides the capacity to achieve high performance
PLATFORM DESIGN AND IMPLEMENTATION SYSTEM OVERVIEW PEARL uses commodity multicore CPU hardware platforms that run generic software as well as specialized packet processing cards for highperformance packet processing as shown in Fig. 1. The virtualization environment is built using the Linux-based LXC solution. This enables multiple virtual router instances to run in parallel over a CPU core or one router instance over multiple CPU cores. Each virtual machine can be logically viewed as a separate host. The hardware platform contains a field programmable gate array (FPGA)-based packet processing card with embedded TCAM and SRAM. This card enables fast packet processing and strong isolation. Isolation — PEARL implements multiple simultaneous fast virtual data planes by allocating separate hardware resources to each virtual data plane. This facilitates strong isolation among the hardware virtual data planes. Moreover, LXC takes advantage of a group of the kernel feature (namespace, Cgroup) to ensure isolation in software between virtual router instances. A multistream high-performance DMA engine is also used in PEARL, which receives and transmits packets via high-speed PCI Express bus between hardware and software platforms. Each IO stream can be either assigned to a dedicated virtual router or shared by several virtual routers using a net IO proxy. Flexibility — We use TAP/TUN device as the network interface in each virtual machine. Each virtual machine could be considered as a stan-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Routing table lookup VR 0 routing table
Action logic 1
MAC RxQ
VR 1 routing table .. .
Action logic 2 .. .
MAC RxQ
CPU transceiver
CPU action
MAC RxQ MAC RxQ
Header extractor
VID lookup
The operations of
Action logic
routing table lookup and packet dispatch
MAC TxQ
Output queues
Out port scheduler
machines are always the performance
MAC TxQ MAC TxQ
to different virtual
Multistream DMA Rx engine PCIE bus
Mapping 10-tuple to virtual router ID
F
Multistream DMA Tx engine
bottleneck. PEARL offloads these two operations into hardware to achieve high speed packet forwarding.
MAC TxQ
Figure 2. The architecture of PEARL's hardware data plane.
dard Linux host containing multiple network ports. Thus, the IPv4, IPv6, OpenFlow, and even non-IP protocol stacks can easily be loaded. Adding new functions to a router is also convenient though programming Linux applications. For example, to load IPv4 or IPv6, the Quagga routing software suite can be used as the control plane inside each Linux container.
can classify packets of any kind of protocol into different virtual routers as long as they are Ethernet based, such as IPV4, IPv6 and non-IP protocols. The VID lookup table is managed by a software controller which enables users to define the fields as needed. The VID of the virtual router to which a packet belongs is appended on the packet as a custom header.
High Performance — The operations of routing table lookup and packet dispatch to different virtual machines are always performance bottlenecks. PEARL offloads these two operations into hardware to achieve high-speed packet forwarding. In addition, since LXC is a lightweight virtualization technique with low overhead, the performance is further improved.
Routing Table Lookup — In a network virtualization environment, each virtual router should have a distinct routing table. Since there are no standards for non-IP protocols until now, we only consider the storage and lookup of routing tables for IP-based protocols in hardware. It is worth noting that routing tables for non-IP protocols can be accommodated through FPGA in the cards. Given limited hardware resources, we implement four routing tables in the current design. The tables are stored in TCAM as well. We take the VID combined with the destination IP address as a search key. The VID part of the key is performed exact matching and the IP part is performed the longest prefix matching in TCAM. Once a packet matches in the hardware, it needn’t be sent to the kernel for further processing, greatly improving the packet forwarding performance. For non-IP protocols or the IP-based protocols that are not accommodated in the hardware, we integrate a CPU transceiver module. The module is responsible for transmitting the packet to the CPU directly without looking up routing tables. Whether a packet should be transmitted by the CPU transceiver module is completely determined by software. With the flexibility offered by the CPU transceiver module, it is easy to integrate more flexible software virtual data planes into PEARL.
HARDWARE PLATFORM To provide both high performance and strong isolation in PEARL, we design a specialized packet processing card. Figure 2 shows the architecture of hardware data plane. It is a pipelinebased architecture, which consists of two main data paths: transmitting and receiving. The receiving data path is responsible for processing the ingress packets, and the transmitting data path processes the egress packets. In the following, the processing stages of the pipeline are detailed. Header Extractor — For each incoming packet, one or many fields are extracted from the packet header. These fields are used for virtual router ID (VID) lookup in the next processing stage. For IP-based protocols, a 10-tuple, defined following OpenFlow [3], is extracted, while for nonIP protocols, the MAC address is extracted. VID Lookup — Each virtual router in the platform is marked by a unique VID. This stage classifies the packet based on the fields extracted in the previous stage. We use TCAM for the storage and lookup of the fields, which can be configured by the users. Due to the special features of TCAM, each field of the rules in a VID lookup table can be a wildcard. Hence, PEARL
Action Logic — The result of routing table lookup is the next hop information, including output card number, output port number, the destination MAC address, and so on, which is stored in an SRAM-based table. It defines how to process the packet, so it can be considered as
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
73
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
High-priority virtual router
The DMA engine can transfer packets to the pre-allocated
New protocols
Routed
A
BEMaGS F
Low-priority virtual router
Physical machine vmmd
huge static buffer at naed
contiguous memory locations. It greatly
User space
low_proxy
Applications
Plug-in netio_proxy
decreases the number of memory accesses required
Kernel space
veth0..4
tap0
macvlan
veth0..4
to transfer a packet to CPU.
Driver
DMA Tx/Rx DMA Tx/Rx
Registers
DMA Tx/Rx
Network cards
Figure 3. Packet path in the software.
an action associated with each entry of the routing tables. Based on the next hop information, this stage performs some decisions such as forwarding, dropping, broadcasting, decrementing time to live (TTL), or updating MAC address. Multi-Stream DMA Engine — To accelerate the IO performance and greatly exploit the parallel processing power of a multicore processor, we design a multistream high-performance DMA engine in PEARL. It can receive packets of different virtual routers from the network card to different memory regions in the host, and transmit packets in the opposite direction via highspeed PCI Express bus. From a software programmer’s perspective, there are multiple independent DMA engines, and the packets of different virtual routers are directed into different memory regions, which is convenient and lockless for programming. Meanwhile, we make a trade-off between flexibility and high performance of DMA transfer mechanism, and carefully redesign the DMA engine in FPGA. The DMA engine can transfer packets to the preallocated huge static buffer at contiguous memory locations. It greatly decreases the number of memory accesses required to transfer a packet to CPU. Each packet transported between the CPU and the network card is equipped with a custom header, which is used for carrying processing information to the destination, such as the matching results. Output Scheduler — The egress packets sent back by the CPU are scheduled based on their output port number, which is a specific field in the custom header of the packet. Each physical output port is equipped with an output queue. The scheduler puts each packet in the appropriate output queue for transmitting.
SOFTWARE PLATFORM Our software platform of PEARL consists of several components, as shown in Fig. 3. These include vmmd, to provide the basic management functions for the virtual routers and packet processing cards; nacd, to offer a uniform interface
74
Communications IEEE
to the underlying processing cards outside the virtual environment; routed, to translate the forwarding rules generated by the kernel or user applications into a uniform format in each virtual router, and install these rules into the TCAM of the processing cards; netio_proxy, to transmit the packets between the physical interfaces and virtual interfaces, and low_proxy, to dispatch packets into low-priority virtual routers that share one pair of DMA Rx/Tx buffers. With different configurations and combinations of these programs, PEARL can generate different types of virtual routers to achieve flexibility. There are two types of virtual routers in our PEARL: high- and low-priority virtual routers. Each high-priority virtual router is assigned one pair of DMA Rx/Tx buffers and an independent TCAM space for lookup. With the high-speed lookup based on TCAM, and efficient IO channels provided by the hardware, the virtual router can achieve the maximum throughput in PEARL platform. For low-priority virtual routers, all the virtual routers share only one pair of DMA Rx/Tx buffers and they can’t utilize TCAM for lookup. The macvlan mechanism is adopted to dispatch packets between multiple virtual routers. The applications inside the low priority virtual routers can use the socket application programming interfaces (APIs) to capture packets from the virtual interfaces. Note that each packets needs to go through a context switch (system calls) at least two times during transmission to the user application, resulting in relatively low IO performance. We take IPv4 and IPv6 as two Internet protocols to show how the virtual routers can easily be implemented on PEARL. High-Priority IPv4 Virtual Router — To create a high-priority IPv4 virtual router in our platform, the vmmd process first starts a new Linux container with several virtual interfaces, and collects the MAC address of each virtual interface, installs these addresses in the underlying cards via the nacd process so that the hardware can identify the packets heading to this virtual router, and copy the packet into the cer-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
tain DMA Tx buffer which assigned to this virtual router. Then, the routed process is launched in the new container. It extracts the routes through the NETLINK socket inside the virtual router and installs routes and the forwarding action in hardware, so the hardware can fill a little structure in the memory to notify the netio_proxy process when a packet match a route in TCAM. The netio_proxy process delivers the packets either to the virtual interface or directly to the physical interface according to the forwarding action in memory. For example, most of time, normal packets will match a route in the hardware. When the netio_proxy receive these packets, it will directly send them through a DMA Tx buffer. An ARP request packet will not match any rules in the TCAM, the netio_proxy process will deliver this packet to the virtual interface, receive the corresponding ARP reply packet from the virtual interface, and then send it to the physical interface. Low-Priority Virtual Router — To create lowpriority virtual routers, a tap device is set up by the vmmd process (tap0). Low priority virtual routers are configured to share the network traffic of this tap device through the macvlan mechanism. The low_proxy process acts like a bridge, transmitting packets between DMA buffers and the tap device in both directions. As the MAC addresses of the virtual interfaces are generated randomly, we can encode the MAC addresses to identify the virtual interface from which a packet comes. For example, we can use the last byte of the MAC address to identify the virtual interfaces,: if the low_proxy process receives a packet with source MAC address 02:00:00:00:00:00, it knows that the packet is from the first virtual interface in one of the low-priority virtual routers, and transmits the packet to the first physical interface immediately. We adopted this method in the low_proxy and vmmd processes, and use the second byte to identify the different low virtual routers. It not only saves the time consumed by the inefficient MAC hash lookup to determine where the packet comes from, but also saves the space in TCAM, because all the low priority virtual routes only need one rule in TCAM (02:*:00:00:00:*). IPv4/IPv6 Virtual Router With Kernel Forwarding — In this configuration, the routed process does not extract the route from the kernel routing table; instead, it enables the ip_forward options of the kernel. As a result, all packets will match the default route in TCAM without the forwarding action. The netio_proxy process transmits all these packets into the virtual interfaces so that the kernel will forward the packet instead of the underlying hardware. The tap/tun device is used as the virtual interface. Since the netio_proxy is a user space process, each packet needs two system calls to complete the whole forwarding. User-Defined Virtual Router — User-define packet process procedure can be implemented as a plug-in loaded by the netio_proxy process, which makes the PEARL extensible. We opened
6
IEEE
BEMaGS F
64 bytes 512 bytes 1518 bytes
4
3
2
1
0
1
2
3
4
Number of VRs
Figure 4. Throughput of high performance IPv4 virtual routers.
the basic packet APIs to users, such as read_packet(), send_packet(). Users can write their own process module in C and run it in the independent virtual routers. For lightweight applications that do not need to deal with huge amounts of network traffic, users can also write a socket program in either high- or low-priority virtual routers.
EVALUATION AND DISCUSSION We implemented a PEARL prototype using a common server with our specialized network card. The common server is equipped with a Xeon 2.5GHz 64-bit CPU and 16G DDR2 RAM. The OS-level virtualization technique Linux Containers (LXC) is used to isolate the different virtual routers (VRs). In order to demonstrate the performance and flexibility of PEARL, our implementation is evaluated in three different configurations: a high-performance IPv4 VR, a kernel-forwarding IPv4/IPv6 VR, and IPv4 forwarding in a low-priority VR. We conducted four independent subnetworks with Spirent TestCenter to measure the performance of the three configurations in one to four VRs. Three different packet lengths (64, 512, and 1518) were used (for IPv6 packet, the packet length is 78, 512, and 1518.; 78 bytes is the minimal IPv6 packet length supported by TestCenter). Figure 4 shows the throughputs of increasing numbers of VRs using configuration 1. Each virtual data plane has been assigned a pair of DMA Rx/Tx buffers, and the independent routing table space in the TCAM, resulting in efficient IO performance and high-speed IP lookup. The result shows, when the number of VRs reaches 2, the throughput of the minimal IPv4 packet of PEARL is up to 4 Gb/s, which is the maximum theatrical speed of our implementation.
IEEE Communications Magazine • July 2011
Communications
A
5
Throughput (G/ps)
Communications
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
75
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Figure 5 illustrates the throughputs of increasing numbers of VRs using configuration 2. Each virtual data plane has the same configuration as configuration 1, except that no virtual data plane has its own routing space in TCAM. In Fig. 5, the throughput of minimal IPv4/IPv6 packet forwarding is only 100 Mb/s when there is only one VR. This is because we used the original kernel for packet forwarding, so each packet needs to go through the context switch twice and memory copy three times in our design. We can optimize this by re-implementing the forwarding functions as a plug-in in the netio_proxy process in VRs. Figure 6 show the throughputs of increasing numbers of low-priority VRs using configuration
A
BEMaGS F
3. Low-priority VRs shares only one pair of DMA TX/RX IO channels and cannot take advantage of TCAM to speed up the IP lookup. It can be used to verify the applications that handle little traffic (new routing protocols etc.). We can see from the results that the total throughput of the minimal IPv4 packet remains 60 Mb/s as the number of VRs increases. The macvlan mechanism used for sharing the network traffic between multiple VRs results in a long kernel path for packet processing, so the total performance is even lower than the IPv4 kernel forwarding in configuration 2. We can improve the performance by developing a new mechanism that suits our case.
RELATED WORK 4 IPv4 64 byte IPv4 512 byte IPv4 1518 byte IPv6 78 byte IPv6 512 byte IPv6 1518 byte
3.5
Throughput (Gb/ps)
3
2.5 2
1.5
1 0.5
0
0
1
2 3 Number of kernel forwarding VRs
4
Figure 5. Throughput of kernel forwarding IPv4/IPv6 virtual routers.
3 64 bytes 512 bytes 1518 bytes 2.5
Throughput (G/ps)
2
1.5
1
0.5
0
0
1
2 3 Number of low-priority VRs
Figure 6. Throughput of low-priority IPv4 virtual routers.
76
Communications IEEE
4
Recent research, such as the vRouter Project [4], RouteBricks [5] and PacketShader [6], have exploited the tremendous power of modern multicore CPUs and multiqueue network interface cards (NICs) to build high-performance software routers on commodity hardware. Memory latency or IO performance becomes the bottleneck for small packets in such platforms. Due to the complexity of the DMA transfer mechanism in commodity NICs, the performance of the highspeed PCI Express bus is not fully exploited. Meanwhile, traditional DMA transfer and routing table lookup result in multiple memory accesses. In PEARL, we simplify the DMA transfer mechanism and redesign the DMA engine to accelerate IO performance, and offload the routing table lookup operation to the hardware platform. OpenFlow [3] enables rapid innovation of various new protocols, and divides the function into control plane and data plane. OpenFlow provides virtualization through flowvisor, without isolation in hardware. In contrast, PEARL hosts multiple virtual data planes in the hardware itself, which could offer both strong isolation and high performance. SwitchBlade [7] builds a virtualized data plane in an FPGA-based hardware platform. It classifies packets based on MAC address. PEARL, on the other hand, dispatches packets into different VRs based on the 10-tuple. With the flexibility of the 10-tuple and wildcard, PEARL has the capability to classify packets based on the protocol of any layer. Moreover, in PEARL, all packets are transmitted to the CPU for switching. The Supercharged Planetlab Platform (SPP) [8] is a high-performance multi-application overlay network platform. SPP utilizes the network processor (NP) to enhance the performance of traditional Planetlab [9] nodes. SPP divides the whole data plane functions into different pipeline stages on NP. Some of these pipeline stages can be replaced by custom code, resulting in extensible packet processing. However, the flexibility and isolation of SPP is limited due to the inherent vendor-specific architecture of NP. PEARL takes advantage of the flexibility and full parallelism of FPGA to run multiple highperformance virtual data planes in parallel, while keeping good isolation in a fast data path. CoreLab [10] is a new network testbed archi-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
tecture supporting a full-featured development environment. It enables excellent flexibility by offering researchers a fully virtualized environment on each node for any arrangement. PEARL can support similar flexibility in the software data path. Meanwhile, PEARL offers a high-performance hardware data path for performance-critical virtual network applications.
CONCLUSIONS We aim at flexible routers to bridge the gap between new Internet protocols and practical test and deployment. To this end, this work presents a programmable virtual router platform, PEARL. The platform allows users to easily implement new Internet protocols and run multiple isolated virtual data planes concurrently. A PEARL router consists of a hardware data plane and a software data plane with DMA engines for packet transmission. The hardware data plane is built on top of an FPGA-based packet processing card with TCAM embedded. The card facilitates fast packet processing and IO virtualization. The software plane is built by a number of modular components and provides easy program interfaces. We have implemented and evaluated the virtual routers on PEARL.
ACKNOWLEDGMENT This work was supported by the National Basic Research Program of China under grant no. 2007CB310702, the National Natural Science Foundation of China (NSFC) under grant no. 60903208, the NSFC-ANR Agence Nationale de Recherche, France) under grant no. 61061130562, and the Instrument Developing Project of the Chinese Academy of Sciences under grant no. YZ200926.
REFERENCES [1] V. Jacobson et al., “(PARC) Networking Named Content,” ACM CoNEXT 2009, Rome, Dec. 2009. [2] M. Handley, O. Hodson, E. Kohler, “XORP: an Open Platform for Network Research,” ACM SIGCOMM Comp. Commun., vol. 33, issue 1, Jan. 2003. [3] N. McKeown et al., “OpenFlow: Enabling Innovation in Campus Networks,” Comp. Commun. Rev., vol. 38, Apr. 2008, pp. 69–74. [4] N. Egi et al., “Towards High Performance Virtual Routers on Commodity Hardware,” Proc. 2008 ACM CoNEXT Conf., Madrid, Spain, 2008. [5] M. Dobrescu et al., “RouteBricks: Exploiting Parallelism to Scale Software Routers,” Proc. ACM SIGOPS 22nd Symp. Op. Sys. Principles, Big Sky, Montana, 2009. [6] S. Han et al., “PacketShader: a GPU-Accelerated Software Router,” ACM SIGCOMM Comp. Commun. Rev., vol. 40, 2010, pp. 195–206. [7] M. Anwer et al., “SwitchBlade: A Platform for Rapid Deployment of Network Protocols on Programmable Hardware,” ACM SIGCOMM Comp. Commun. Rev., vol. 40, 2010, pp. 183–94.
[8] J. Turner et al., “Supercharging PlanetLab — A High Performance, Multi-Application, Overlay Network Platform,” ACM SIGCOMM ’07, Kyoto, Japan, Aug. 27–31, 2007. [9] B. Chun et al., “Planetlab: an Overlay Testbed for Broad-Coverage Services,” ACM SIGCOMM ’03, Karlsruhe, Germany, Aug. 25–29, 2003. [10] CoreLab, http://www.corelab.jp/
IEEE
BEMaGS
PEARL takes flexibility and full parallelism of FPGA
BIOGRAPHIES
to run multiple
G AOGANG X IE (
[email protected]) ________ received a Ph.D. degree in computer science from Hunan University in 2002. He is a professor at the Institute of Computing Technology (ICT), Chinese Academy of Sciences (CAS). His research interests include future Internet architecture, programmable virtual router platforms, and Internet measurement and modeling.
high-performance
P ENG H E (
[email protected]) __________ received his B.S. degree in electronic and information engineering from Huazhong University of Science and Technology in 2004. He is now pursuing a Ph.D. degree in computer architecture at the ICT, CAS. His research interests include high-performance packet processing and virtual router design.
F
advantage of the
virtual data planes in parallel, while keeping good isolation in fast data path.
H ONGTAO G UAN (
[email protected]) ______________ is currently an assistant professor of ICT, CAS. His research interests include visualization technology of computer networks and routers. He studied computer science at Tsinghua University from 1999 to 2010 and obtained B.E. and Ph.D. degrees. He took part in the BitEngine12000 project from 2002 to 2005, which was the first IPv6 core router in China. Z HENYU L I (
[email protected]) ________ received a Ph.D. degree from ICT/CAS in 2009, where he serves as an assistant professor. His research interests include future Internet design, P2P systems, and online social networks. YINGKE XIE (
[email protected]) ________ received his Ph.D. degree from ICT, CAS, where he serves as an associate professor. His research interests include high-performance packet processing, reconfigurable computing, and future network architecture. L AYONG L UO (
[email protected]) ____________ is currently working toward his Ph.D. degree at ICT, CAS. He is currently a research assistant in the Future Internet Research Group of ICT. His research interests include programmable virtual router, reconfigurable computing (FPGA), and network virtualization. JIANHUA ZHANG (
[email protected]) _____________ received his B.S. and M.S. from University of Science & Technology Beijing, China, in 2006 and 2009, respectively. He is currently a Ph.D. student at ICT, CAS. His research interest is in the future Internet. YONGGONG WANG (
[email protected]) ______________ received his B.S. and M.S. from Xidian Univiserty, Xi’an, China, in 2005 and 2008, respectively. He is currently a Ph.D. student at ICT, CAS. His research interest is in the future Internet. KAVÉ SALAMATIAN (
[email protected]) _______________ is a professor at the University of Savoie. His main areas of research are Internet measurement and modeling and networking information theory. He was previously a reader at Lancaster University, United Kingdom, and an associate professor at Université Pierre et Marie Curie. He graduated in 1998 from Paris SUD-Orsay University, where he worked on joint source channel coding applied to multimedia transmission over Internet for his Ph.D. He is currently a visiting professor at the Chinese Academy of Science.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
77
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
SERIES EDITORIAL
TOPICS IN NETWORK AND SERVICE MANAGEMENT
George Pavlou
T
his is the 11th issue of the series on Network and Service Management, which is typically published twice a year. It was originally published in April and October but since last year it is published in July and December. The series provides articles on the latest developments in this well-established discipline, highlighting recent research achievements and providing insight into both theoretical and practical issues related to the evolution of the discipline from different perspectives. The series provides a forum for the publication of both academic and industrial research, addressing the state of the art, theory. and practice in network and service management. During the 12th Integrated Management Symposium (IM 2011), which was held in Trinity College Dublin last May, the new CNOM committee was elected after a call for nominations. The previous Chair, Dr. Marcus Brunner of NEC Europe, was re-elected, while the previous Vice Chair, Prof. George Pavlou of University College London, stepped down after having served a term of two years. The previous Technical Program Chair, Prof. Lisandro Granville of the Federal University of Rio Grande, Brazil, was elected Vice Chair, while Dr. Filip de Turck of Ghent University, Belgium, was elected Technical Program Chair. Finally, the previous Secretary, Dr. Brendan Jennings of Waterford Institute of Technology, Ireland, was re-elected to this position. The new Chairs plan to continue the organization of successful events such as IM, NOMS, CNSM, and others. Key to the success of these events has been the good collaboration with the IFIP TC6 sister organization WG 6.6, whose Chair is Prof. Aiko Pras of the University of Twente, Netherlands, with Co-Chair Dr. Olivier Festor of INRIA, France. During the same Symposium (IM 2011) it was announced that Prof. Morris Sloman has been elected IEEE Fellow. Professor Sloman is amongt the most influential researchers in our field, having established and provided seminal contributions in policy-based management, and current Editor-in-Chief of IEEE Transactions on Network and Service Management (TNSM). In addition, Prof. George Pavlou received the Dan Stokesberry Award. This
78
Communications IEEE
Aiko Pras
award is given in memory of IM ’97 Chair Dan Stokesberry at each IM Symposium to an individual who has made particularly distinguished technical contributions to the growth of the field. Professor Pavlou has contributed significantly in various areas, including most notably management protocols and quality of service management. The previous recipients of the Dan Stokesberry award have been Robbie Cohen of AT&T (1997), Morris Sloman of Imperial Collage London (1999), Heinz-Gerd Hegering of Ludwig Maximilian University of Munich (2001), Aurel Lazar of Columbia University (2003), John Strassner of Motorola Research (2005), Joe Hellerstein of Microsoft (2007), and Raouf Boutaba of the University of Waterloo (2009). The European Conference on Autonomous Infrastructure Security & Management (AIMS), which was established by the EMANICS Network of Excellence, continued running this year, with its fifth instance (AIMS 2011) taking place at Nancy, France; http://www.aims-con_______________ ference.org/2011/. The key annual event in this area, which ____________ this year was the 12th IEEE/IFIP Integrated Management Symposium (IM 2011), was held 23–27 May in Dublin, Ireland; http://www.ieee-im.org/2011/. The second key annual event in this area is the relatively new IEEE/IFIP Conference on Network & Service Management (CNSM 2011), which has become another flagship event, complementing IM and NOMS. This year’s second CNSM will take place 2428 October in Paris, France, and aims to repeat the success of last year’s conference in Niagara Falls, Canada; http://www.cnsm2011.fr/. We again experienced overwhelming interest in the 11th issue, receiving 22 submissions in total. For all the articles we got at least three independent reviews each. We finally selected four articles, resulting in an acceptance rate of 18.2 percent. It should be mentioned that the acceptance rate for all the previous issues has ranged between 14 and 25 percent, making this series a highly competitive place to publish. We intend to maintain our rigorous review process in future issues, thus maintaining the high quality of the published articles.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
SERIES EDITORIAL The first article, “Toward Decentralized Probabilistic Management” by Gonzalez Prieto, Gillblad, Miron, and Steinert, argues that the adoption of a decentralized and probabilistic paradigm for network management will be crucial to meet the challenges of future networks and illustrates the paradigm through example solutions of real-time monitoring and anomaly detection. The second article, “Network Resilience: A Systematic Approach” by Smith, Scholler, Fessi, Karaliopoulos, Lac, Hutchison, Sterbenz, and Plattner, presents a systematic approach to building resilient networked systems, studying fundamental elements at a framework level and demonstrating its applicability through a concrete case study. The third article, “A Survey of Virtual LAN Usage in Campus Networks” by Yu, Sun, Feamster, Rao, and Rexford, describes how three university campuses and one academic department use VLANs to achieve a variety of goals, arguing that VLANs are ill suited to some of these goals and that their use leads to significant complexity in the configuration of network devices. Finally, the fourth article, “Toward Fine-Grained Traffic Classification” by Park, Won, and Hong, proposes a fine-grained traffic classification scheme based on the analysis of existing classification methodologies and demonstrates that the proposed scheme can provide more in-depth classification results for analyzing user contexts. We hope that readers of this issue find again the articles informative and we will endeavor to continue with
similar issues in the future. We would finally like to thank all the authors who submitted articles to this series and the reviewers for their valuable feedback and comments on the articles.
BIOGRAPHIES G EORGE P AVLOU (
[email protected]) _____________ is a professor of communication networks in the Department of Electronic and Electrical Engineering, University College London, United Kingdom. He received a Diploma in engineering from the National Technical University of Athens, Greece, and M.Sc. and Ph.D. degrees in computer science from University College London. His research interests focus on network management, networking, and service engineering, including aspects such as traffic engineering and quality of service management, policy-based systems, autonomic networking, content-centric networking, and communications middleware. He has been instrumental in a number of European and U.K. research projects, and has contributed to standardization activities in ISO, ITU-T and IETF. He is Vice Chair of IEEE CNOM, and was the technical program co-chair of the Seventh IFIP/IEEE Integrated Management Symposium (IM 2001) and the Tenth IFIP/IEEE International Conference on Management of Multimedia and Mobile Networks and Services (MMNS 2008). AIKO PRAS (
[email protected]) __________ is an associate professor in the Departments of Electrical Engineering and Computer Science at the University of Twente, the Netherlands, and a member of the Design and Analysis of Communication Systems Group. He received a Ph.D. degree from the same university for his thesis, Network Management Architectures. His research interests include network management technologies, network monitoring, measurements, and security. He chairs IFIP Working Group 6.6 on “Management of Networks and Distributed Systems,” and has been research leader in the European Network of Excellence on Management of the Internet and Complex Services (EMANICS). He is a Steering Committee member of the IFIP/IEEE NOMS, IM, CNSM, and AIMS Symposia, and has been technical program chair of several conferences, including IFIP/IEEE Integrated Management Symposium 2005 (IM 2005) and IFIP/IEEE Management Week 2009 (ManWeek2009).
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
79
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
TOPICS IN NETWORK AND SERVICE MANAGEMENT
Toward Decentralized Probabilistic Management Alberto Gonzalez Prieto, Cisco Systems Daniel Gillblad and Rebecca Steinert, Swedish Institute of Computer Science Avi Miron, Israel Institute of Technology
ABSTRACT In recent years, data communication networks have grown to immense size and have been diversified by the mobile revolution. Existing management solutions are based on a centralized deterministic paradigm, which is appropriate for networks of moderate size operating in relatively stable conditions. However, it is becoming increasingly apparent that these management solutions are not able to cope with the large dynamic networks that are emerging. In this article, we argue that the adoption of a decentralized and probabilistic paradigm for network management will be crucial to meet the challenges of future networks, such as efficient resource usage, scalability, robustness, and adaptability. We discuss the potential of decentralized probabilistic management and its impact on management operations, and illustrate the paradigm by three example solutions for real-time monitoring and anomaly detection.
INTRODUCTION
The work presented in this article was done while the first author was at the Royal Institute of Technology (KTH), Sweden.
80
Communications IEEE
Current network management solutions typically follow a centralized and deterministic paradigm, under which a dedicated station executes all management tasks in order to manage a set of devices [1]. This paradigm has proved successful for communication networks of moderate size operating under relatively stable network conditions. However, the growing complexity of communication networks, with millions of network elements operating under highly dynamic network conditions, poses new challenges to network management, including efficient resource usage, scalability, robustness, and adaptability. Scalability and robustness are essential for coping with the increasing network sizes while providing resilience against failures and disturbances. Adaptability of management solutions will be crucial, as networks to a higher degree will operate under network conditions that vary over time. Furthermore, since communication and computational capacities are limited, it is critical that management solutions use
0163-6804/11/$25.00 © 2011 IEEE
network resources efficiently. In order to meet these challenges, new approaches to network management must be considered. We believe that two key approaches to network management will become increasingly important: decentralized management, where management algorithms operate locally at the managed nodes without centralized control; and probabilistic management, in which management decisions and policies are not based on deterministic and guaranteed measurements or objectives. As we argue in this article, decentralized and probabilistic approaches are well positioned to solve the critical network management challenges listed above, and therefore a move toward decentralized probabilistic management seems likely. This move, however, means a fundamental shift in how network performance is viewed and controlled, and introduces operational changes for those that configure and maintain networks. In this article we discuss decentralized probabilistic management, describing how it can solve pressing challenges that large-scale dynamic networks bring to management, and how the adoption of this approach impacts network operations. We discuss the benefits, drawbacks, and impact of moving toward decentralized probabilistic management. We provide evidence and directions of current developments, while we present three concrete examples that demonstrate merits of decentralized probabilistic management.
DECENTRALIZED PROBABILISTIC MANAGEMENT ASPECTS OF PROBABILISTIC MANAGEMENT Management solutions can be probabilistic in one or several aspects. A probabilistic management algorithm: •Makes use of probabilistic models for representing the network state. That is, the algorithm represents the network state as probability distributions, using, for example, simple parametric representations [2], more complex graphical- or mixture models [3, 4], or sampling from generative models.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
•Does not necessarily use deterministic objectives, but rather objectives that are specified in terms of probabilities and uncertainties [2, 5]. This could, for example, mean specifying a bound on service failure probability. •Might provide its output in terms of probability distributions, instead of deterministic values. As an example, the average link load over a network could be reported with an expected mean value and variance instead of a single value [6]. •Might implement a probabilistic sampling algorithm, say, by randomly turning on and off management functions [6], relying on a random subset of nodes for estimating an aggregate value for all nodes [7], or sampling relevant network parameters at random rather than regular intervals [8]. •Might perform network control actions or explore the consequences of control actions using an element of randomness [9]. For example, the efficiency of different routing strategies could be continuously explored to adapt to changing network conditions. The first of these aspects is inherent to all probabilistic management solutions, whereas the other aspects listed may or may not be present in a specific algorithm.
POTENTIAL BENEFITS AND DRAWBACKS Decentralized approaches have several benefits: they scale very well with increasing network size, adapt to churn well, avoid single points of failure, and thus can be significantly more robust than centralized solutions. Probabilistic approaches can be significantly more resource-efficient than deterministic ones. By using the slack provided by the use of probabilistic rather than deterministic objectives (which often assume worst-case scenarios), the amount of bandwidth and processing resources consumed by the solutions can be significantly reduced. This feature is highly valuable for volatile networks, such as in cloud computing. Probabilistic approaches are also highly suitable for efficiently managing uncertainty and noise, thereby improving robustness in algorithm performance. Furthermore, probabilistic management changes the way networks are configured, as goals of the management algorithms can be stated as acceptable probabilities of distributions of performance metrics, allowing for intuitive interaction with the operator. A possible drawback of decentralized approaches is that the lack of centralized control typically leads to suboptimal solutions to management problems. For probabilistic approaches a possible drawback follows from the introduction of uncertainty. It does not provide operators with deterministic control over the state of managed devices. Note that while probabilistic management cannot provide hard guarantees on the accuracy of network measurements, capturing an accurate snapshot of today’s dynamic and volatile networks is already impossible. Managers must already accept some level of uncertainty in the data collected for management operations, and the introduction of probabilistic approaches might therefore come with acceptable cost in this regard.
The combined benefits of decentralized and probabilistic approaches amount to management solutions that operate in a failure-resilient and resource-efficient manner. The potential drawbacks are less detailed control over managed devices and suboptimal solutions.
IEEE
BEMaGS F
Network managers will need to think about the network in terms of uncer-
IMPACT ON NETWORK MANAGEMENT
tainties and likely
The introduction of decentralized probabilistic management approaches will bring a paradigm shift in how we specify network objectives and view network performance. Specifically, terms such as risk and uncertainty must now be taken into account when configuring and running networks. For example, although probabilistic management solutions may achieve better performance on average, they may, with some small probability miss objectives or perform badly. Managers can no longer rely on strict guarantees on network performance, and need to view the goals of the network as an expected service quality while allowing for some variance. Additionally, the information that management and configuration decisions are based on is no longer guaranteed to be within deterministic bounds, and could, with a low probability, be out of the expected range. With the new paradigm, network managers will not only be utilizing strict rules, but will also be specifying objectives that include uncertainties. In addition to monitoring network performance, the role of a network manager is likely to also include monitoring the performance of decentralized probabilistic management solutions. Related to this, managers will have to analyze problems in the network differently, as the increased autonomy and uncertainty of decentralized probabilistic management solutions could potentially allow for some underlying problems to remain undetected. In summary, network managers will need to think about the network in terms of uncertainties and likely scenarios while considering the expected cost and cost variability. This is a very different approach from current practices, where managers focus on upholding strict guarantees on parameters staying within predetermined limits.
scenarios while considering the expected cost and cost variability. This is a very different approach from current practices.
FURTHER CONSIDERATIONS Autonomous mechanisms are an important part of efficient management processes in complex networks. Naturally, adaptability is key to achieving the autonomy necessary to reduce operative costs in future networks. Although facilitated by a probabilistic approach, adaptability is by no means an inherent property and must be considered during method development. However, the combination of decentralized and probabilistic approaches allows for the design of highly adaptable solutions. Increased autonomy and adaptability will likely lead to less detailed control of managed resources for the network operators. In our view, this means that decentralized probabilistic solutions need to be developed with an additional property in mind to gain acceptance: the solutions must provide managers with an accurate prediction of the performance of a management solution (e.g., the amount of resources it con-
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
81
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Probabilistic practices are already widely used in various networking and communications processes. Examples include CSMA/CD, the Ethernet protocol ensuring that only one node is transmitting its data on the network wire at any point of time; and statistical multiplexing in IP networks for better resource efficiency.
82
Communications IEEE
sumes), and managers need to be able to control this performance. As illustrated by specific examples provided later, this type of performance control and prediction is actually rather straightforward to express in the adaptive probabilistic models we envision.
CURRENT DEVELOPMENT This section presents prior work in the area of decentralized probabilistic management systems, discussing, in turn, the management framework, and management operations in the area of monitoring, diagnosis and traffic management. Probabilistic approaches can be used to enhance the management infrastructure in terms of resource efficiency [6, 10]. A probabilistic decentralized framework for network management is presented in [6], in which management functions are randomly turned on or off, thereby effectively exploiting redundancy in those functions. By means of simulation, the study demonstrates reduced effort and resources required for performance and fault management operations, while still achieving a sound level of accuracy in the overall network view. Relying on probabilistic estimates, self-organization of sensor and ad hoc networks is discussed in [10]. The study makes use of connectivity probability information in order to select the management clusters that can efficiently carry out the management tasks. A further example makes use of random sampling among network entities, where network monitoring operations benefit from gossip-based solutions for aggregation of network management data. This can be carried out by nodes that collaborate with randomly selected neighbors [11]. Neighbor sampling during neighbor discovery results in more efficient data dissemination than non-gossip flooding schemes. A more recent study [12] further improves the efficiency of the neighbor discovery process by implementing a probabilistic eyesight direction function, which practically narrows the direction through which neighbors are sought. Network diagnosis is explored in multiple studies using probabilistic representations such as graphical models to infer the fault. Two studies also add decentralized processing [3, 4]. Moreover, [4] implements a collaborative approach for a Bayesian network model. The scheme effectively handles uncertainty in dynamic networks, which was demonstrated on three different scenarios. The probabilistic approach makes it possible to provide diagnostics with a limited amount of information, although the higher the amount of evidencs, the greater certainty the system gets. The study reported in [3] presents an extensible Bayesian-inference architecture for Internet diagnosis, which deploys distributed diagnostic agents and includes a component ontology-based probabilistic approach to diagnosis. The proposed architecture was successfully demonstrated with realworld Internet failures over a prototype network. Traffic management operations can also take advantage of probabilistic approaches. In [8], the authors propose sampling the data such that only a small number of packets are actually cap-
A
BEMaGS F
tured and reported, thereby reducing the problem to a more manageable size. End user congestion control mechanisms are presented in [13], which interact with a probabilistic active queue management of flows. The model captures the packet level dynamics and the probabilistic nature of the marking mechanism for investigating the bottleneck link and profiling the queue fluctuations, eventually gaining better understanding regarding the dynamics of the queue, and means to cope with congestion. The sample studies presented above demonstrate preliminary positive experience gained with decentralized probabilistic approaches for network management, and exemplifies directions for further evolution toward decentralized probabilistic management. Probabilistic practices are already widely used in various networking and communications processes. Examples include carrier sense multiple access with collision detection (CSMA/CD), the Ethernet protocol ensuring that only one node is transmitting its data on the network wire at any point of time; and statistical multiplexing in IP networks for better resource efficiency. These examples cope with vast amounts of information and limited resource availability by implementing probabilistic methods in a distributed manner. There is no question that such probabilistic practices are successful, as shown by the fact that they are standardized and widely deployed. As networks increase in scale, it is likely that network management operations will also start deploying decentralized probabilistic approaches. The studies reported above represent early examples in such a development.
PROBABILISTIC MANAGEMENT ALGORITHMS In this section we present three algorithms to illustrate different aspects of decentralized probabilistic management and applications within fault and performance monitoring. Such algorithms are responsible for estimating the network state, and are crucial functions of a complete management system as they support other management tasks, including fault, configuration, accounting, performance, and security management. First, we discuss a probabilistic approach to anomaly detection and localization, which makes use of local probabilistic models to adapt to local network conditions, while setting the management objectives in probabilistic terms. Second, we describe a tree-based algorithm for probabilistic estimation of global metrics, in which both objectives and reported results are probabilistic in nature. Finally, we present an alternative scheme for the estimation of network metrics by counting the number of nodes in a group, which makes use of random sampling over network entities in addition to the use of probabilistic objectives and outputs.
PROBABILISTIC ANOMALY DETECTION We have devised a distributed monitoring algorithm that provides autonomous detection and localization of faults and disturbances, while
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
adapting its use of network resources to local network conditions [2, 14]. The decentralized approach is based on local probabilistic models created from monitored quality of service (QoS) parameters, such as drop rate and link latency, measured by probing. Based on the probabilistic models, the distributed algorithm provides link disturbance monitoring and localization of faults and anomalies. The anomaly detection approach serves as an example for some of the different aspects of using probabilistic management. The algorithm operates based on probabilistic objectives as input, instead of deterministically set parameters. This significantly reduces the requirements on manual configuration, of either individual network components or across the network, even when network conditions are highly variable. Given such probabilistic objectives, estimated probability models are used for adjusting relevant low-level algorithm parameters. The autonomous adjustment of low-level parameters matching the set of probabilistic objectives enables predictive control of the algorithm performance within probabilistic guarantees. Moreover, the probabilistic models can be used for prediction and decision making, as estimated model parameters can be extracted from the algorithm. This enables other parts of the network management (e.g., traffic management) to take advantage of the estimated probabilistic models for autonomous configuration of network parameters. To run the algorithm, the managing operator specifies a number of high-level management requirements, in terms of network resources, to run the algorithm. Here, the high-level requirements are expressed as probabilities related to probing traffic and detection delays. Low-level parameters, such as probing rates and probing intervals, autonomously adapt to current network conditions and the management requirements. Specifically, the operator sets the acceptable fraction of false alarms in detected anomalies. Moreover, the operator specifies a fraction of the estimated probability mass of observed probe response delays on a link [2]. Thereby, accuracy and probing rates are specified as probabilities rather than being specified in terms of a fixed probing rate across the entire network, providing a typical example of configurations expressed as probabilistic goals rather than in terms of deterministic limits. The use of these probabilistic parameters effectively determines the normally observed probing rates for each link and how quickly action is taken to confirm a suspected failure, while losses and delays are accounted for [2]. Figure 1 depicts examples of the algorithm behavior for one network link. We observe that the obtained rate of false alarms successfully meets the management objective of the specified acceptable false alarm rate for different rates of packet drop (Fig. 1a). In fact, the acceptable rate of false alarms is here an upper limit of the expected amount of false positives on an individual link, which exemplifies the difference between strict performance guarantees and predictive performance control, when using proba-
Rate (log. scale)
IEEE
IEEE
BEMaGS F
-3 -5 -3
-2.5 -2 -1.5 Fraction of acceptable false alarms (log. scale)
-1
(a) 50 30 10 -3.5
-3
-2.5 -2 -1.5 Fraction of acceptable false alarms (log. scale)
-1
(b) Drop=0.2
Drop=0.3
Drop=0.4
Drop=0.5
Figure 1. a) Rate of false alarms given a fraction of acceptable false alarms and drop; b) adaptive probe rates given a fraction of acceptable false alarms and drop. bilistic management algorithms. Similarly, we show that the number of probes needed for detecting a failure adapts to the observed network conditions in order to meet the same requirements on the rate of acceptable false alarms (Fig. 1b), exemplifying how the estimated probabilistic models can be used for autonomous adjustments of low-level parameters.
TREE-BASED PROBABILISTIC ESTIMATION OF GLOBAL METRICS Accuracy — Generic Aggregation Protocol (AGAP) is a monitoring algorithm that provides a management station with a continuous estimate of a global metric for given performance objectives [5]. A global metric denotes the result of computing a multivariate function (e.g., sum, average, and max) whose variables are local metrics from nodes across the networked system (e.g., device counters or local protocol states). Examples of global metrics in the context of the Internet are the total number of VoIP flows in a domain or the list of the 50 subscribers with the longest end-to-end delay. A-GAP computes global metrics in a distributed manner using a mechanism we refer to as in-network aggregation. It uses a spanning tree, whereby each node holds information about its children in the tree, in order to incrementally compute the global metric. The computation is push-based in the sense that updates of monitored metrics are sent toward the management station along the spanning tree. In order to achieve efficiency, we combine the concepts of in-network aggregation and filtering. Filtering drops updates that are not significant when computing a global metric for a given accuracy objective, reducing the management overhead. A key part of A-GAP is the model it uses for the distributed monitoring process, which is based on discrete-time Markov chains. The model allows us to describe the behavior of individual nodes in their steady state and relates
IEEE Communications Magazine • July 2011
Communications
A
-1
-3.5
Rate
Communications
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
83
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
300
Overall updates/s
250 200 150 100 50 0 0
10
20
30
40
50
Aggregating nodes
Figure 2. Overall management overhead as a function of the number of aggregating nodes (for a network with 200 nodes). performance metrics to control parameters. The model has been instrumental in designing a monitoring protocol that is controllable and achieves given performance objectives. The managing operator specifies the global metric of interest and then the desired accuracy as a probability distribution for the estimation error. Examples of input A-GAP supports include the average error, percentile errors, and maximum error. As a consequence, our design permits any administrator with a basic understanding of performance metrics to use our solution, without the need for detailed knowledge of our solution internals. Based on this input, the algorithm continuously adapts its configuration (e.g., filters), providing the global metric with the required accuracy. The output of the algorithm is a continuous estimate of the global metric with the required accuracy. The output also includes predictions for the error distribution and traffic overhead. This output is provided in real time at the root node of the spanning tree. A-GAP has proved to make an efficient use of resources. Figure 2 is a representative example of A-GAP’s performance: its maximum overhead increases sublinearly with the network size (for the same relative accuracy). In addition, the overall management overhead scales logarithmically with the number of internal (i.e., aggregating) nodes in the spanning tree. This behavior is consistent in all our experiments, where we have used both synthetic and real traces, and a wide range of network topologies [5]. Our experiments include both simulation and testbed implementations.
PROBABILISTIC ESTIMATION OF GROUP SIZES Not All at Once! (NATO!) is a probabilistic algorithm for precisely estimating the size of a group of nodes meeting an arbitrary criterion without explicit notification from every node [7]. The algorithm represents an example of a probabilistic sampling approach and provides an alternative to aggregation techniques. It can be used to collect information such as the number of nodes with high packet rate, indicating emerging congestion. By not having
84
Communications IEEE
A
BEMaGS F
each node reporting its above-normal metrics independently, available capacity and resources are efficiently utilized, thereby avoiding excessive amounts of traffic at the ingress channel of the management station. The scheme provides control over the tradeoffs between data accuracy, the time required for data collection, and the amount of overhead incurred. NATO! is an example of a family of algorithms that implement probabilistic polling for estimating the size of a population. It implements a distributed scheme in which nodes periodically and synchronously send reports only if their metrics exceed a threshold after waiting a random amount of time sampled from an agreed time distribution function. The network management station waits until it receives a sufficient number of reports to estimate the total number of nodes with the desired precision, and broadcasts a stop message, notifying the nodes that have not yet reported not to send their reports. The management station then analyzes the transmission time of the received reports, defines a likelihood function, and computes the number of affected nodes for which the likelihood function is maximized. Typically, with only 10 report messages coming from a group of 1000 or 10,000 nodes, the estimation error is practically eliminated. This significant reduction in network load is achieved at the expense of marginal computation load at each node and the broadcast messages. The scheme is an effective monitoring platform that demonstrates efficient resource usage, scalability, robustness, adaptability, autonomy, and ease of use. Network managers control and configure NATO! by means of a number of high-level parameters: the desired metrics to monitor and their threshold values, acceptable overhead (specified as the maximum allowed rate of incoming messages), the time it takes to conclude the number of nodes experiencing an abnormal condition, and the desired accuracy of the estimation. A simple heuristic translates these parameters to a specific time distribution function and a time interval, and the frequency at which NATO! is implicitly invoked. This configuration controls the trade-off between accuracy, timeliness, and overhead. It can be dynamically adapted when the network conditions or management objectives change: faster estimations can be delivered, setting a shorter time interval for the time distribution function, at the expense of higher density of the incoming messages; for faster reaction to changes in network conditions, the frequency at which NATO! is invoked can be increased; when in-depth analysis is required, threshold values of network metrics can be changed and new network metrics can be added; for better fault localization, local NATO! managers can be assigned to collect data in their subnetworks. All of these configuration changes can become active by means of a broadcast message from the management station, which adapts the monitoring task to the current needs for best performance under acceptable cost. Due to its probabilistic nature, the scheme is practically scalable to any network size. There is
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
an insignificant incremental overhead for larger network domains at the egress channel, delivering the broadcast messages from the management station to a larger group of nodes. However, this overhead fans out quickly while following the topology tree, without any detrimental effect on the stressed ingress channel of the management station.
LEVERAGING DECENTRALIZED PROBABILISTIC MANAGEMENT All three algorithms presented in this section demonstrate the benefits of decentralized probabilistic approaches to network management: they enable efficient resource usage compared to deterministic approaches, exploiting currently available resources, while taking noise and variations in the network into account. The algorithms are scalable, robust, and adaptive to network conditions. The decentralized approach for anomaly detection enables scalable and efficient usage of resources. The use of adaptive probabilistic models allows for capturing the local network behavior and predicting the algorithm performance. For A-GAP, decentralization enables efficiency, and probabilistic management provides performance control. Computing metrics in a distributed fashion along an aggregation tree permits reducing the monitoring traffic compared to a centralized approach. The use of probabilistic models permits A-GAP to predict its performance and therefore meet the objectives of the network manager. Thanks to its decentralized and probabilistic nature, NATO! is scalable to any network size, avoiding congestion at the ingress channel of the management station while effectively controlling the trade-off between accuracy, timeliness, and overhead.
CONCLUSIONS In this article we have advocated for the adopt ion o f a de c e nt ra l i z e d an d p r o b ab i l i s t i c paradigm for network management. We have argued that solutions based on it can meet the challenges posed by large-scale dynamic networks, including efficient resource usage, scalability, robustness, and adaptability. We have exemplified this with three specific solutions, and discussed how they address such challenges. A key challenge in the adoption of this paradigm is acceptance by network managers. They will need to think about the network in terms of uncertainties and likely scenarios while considering the expected cost and cost variability. This is a very different approach from current practices, where managers focus on upholding strict guarantees on parameters staying within pre-determined limits. Operators must look at the state of their networks in terms of probabilities and accept a certain degree of uncertainty. Solutions that can quantify that uncertainty are, from this point of view, of great relevance. This paradigm shift is also likely to cause
some reluctance among network operators to deploy this type of solutions. We believe that in order to mitigate this reluctance, it will be key to show that decentralized probabilistic approaches can reduce operational expenditures. This reduction is enabled by their higher degree of automation compared to traditional approaches. However, at this point, this is a conjecture and must be supported by developing use cases that quantify the potential savings in different scenarios. For this purpose, experimental evaluations in large-scale testbeds and production networks are a must. While we expect some reluctance in adopting a new management paradigm, we strongly believe that not doing it would have a major negative impact on the ability to manage larger and more complex networks, and as the need for solutions for such networks increases, we are likely to see more widespread adoption.
IEEE
BEMaGS F
While we expect some reluctance in adopting a new management paradigm, we strongly believe that not doing it would have a major negative impact on the ability to manage larger and more complex networks
ACKNOWLEDGEMENT This work was supported in part by the European Union through the 4WARD and SAIL projects (http://www.4ward-project.eu/, http:// ____ www.sail-project.eu/) in the 7th Framework Programme. The authors would like to thank Reuven Cohen (Technion), Björn Levin (SICS), Danny Raz (Technion), and Rolf Stadler (KTH) for their valuable input to this work.
REFERENCES [1] G. Pavlou, “On the Evolution of Management Approaches, Framework and Protocols: A Historical Perspective,” J. Network and Sys. Mgmt., vol. 15, no. 4, Dec. 2007, pp. 425–45. [2] R. Steinert and D. Gillblad, “Towards Distributed and Adaptive Detection and Localisation Of Network Faults,” AICT 2010, Barcelona, Spain, May 2010. [3] G. J. Lee, CAPRI: A Common Architecture for Distributed Probabilistic Internet Fault Diagnosis, Ph.D. dissertation, CSAIL-MIT, Cambridge, MA, 2007. [4] F. J. Garcia-Algarra et al., “A Lightweight Approach to Distributed Network Diagnosis under Uncertainty,” INCOS ’09, Barcelona, Spain, Nov. 2009. [5] A. Gonzalez Prieto, “Adaptive Real-Time Monitoring for Large-Scale Networked Systems,” Ph.D. dissertation, Dept. Elect. Eng., Royal Insti. Technology, KTH, 2008. [6] M. Brunner et al., “Probabilistic Decentralized Network Management,” Proc. IEEE IM ’09, New York, NY, 2009. [7] R. Cohen and A. Landau, “Not All At Once! — A Generic Scheme for Estimating the Number of Affected Nodes While Avoiding Feedback Implosion,” INFOCOM 2009 Mini-Conf., Rio di Janeiro, Brazil, Apr. 2009. [8] K. C. Claffy, G. C. Polyzos, and H.-W. Braun, “Application of Sampling Methodologies to Network Traffic Characterization,” ACM SIGCOMM Comp. Commun. Rev., vol. 23, no. 4, Oct. 1993, pp. 194–203. [9] E. Stevens-Navarro, L. Yuxia, and V. W. S. Wong, “An MDP-Based Vertical Handoff Decision Algorithm for Heterogeneous Wireless Networks,” IEEE Trans. Vehic. Tech., vol. 57, no. 2, 2008. [10] R. Badonnel, R. State, and O. Festor, “Probabilistic Management of Ad Hoc Networks,” Proc. NOMS ’06, Vancouver, Canada, Apr. 2006, p. 339–50. [11] A. G. Dimakis, A. D. Sarwate, and M. Wainwright, “Geographic Gossip: Efficient Aggregation for Sensor Networks,” IPSN 2006, Nashville, TN, Apr. 2006. [12] L. Guardalben et al., “A Cooperative Hide and Seek Discovery over in Network Management,” IEEE/IFIP NOMS Wksps. ’10 Osaka, Japan, Apr. 2010, pp. 217–24. [13] P. Tinnakornsrisuphap and R. J. La, “Characterization of Queue Fluctuations in Probabilistic AQM Mechanisms,” Proc. ACM SIGMETRICS, 2004, pp. 283–94. [14] R. Steinert and D. Gillblad, “Long-Term Adaptation and Distributed Detection of Local Network Changes,” IEEE GLOBECOM, Miami, FL, Dec. 2010.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
85
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
BIOGRAPHIES ___________ received his ALBERTO GONZALEZ PRIETO (
[email protected]) M.Sc. in electrical engineering from the Universidad Politecnica de Cataluña, Spain, and his Ph.D. in electrical engineering from the Royal Institute of Technology (KTH), Stockholm, Sweden. He has been with Cisco Systems since 2010. He was an intern at NEC Network Laboratories, Heidelberg, Germany, in 2001, and at AT&T Labs Research, Florham Park, New Jersey, in 2007. His research interests include management of large-scale networks, real-time network monitoring, and distributed algorithms.
DANIEL GILLBLAD (
[email protected]) ______ has a background in statistical machine learning and data analysis, and has extensive experience in applying such methods in industrial systems. He holds an M.Sc. in electrical engineering and a Ph.D. in computer science, both from KTH. He has been with the Swedish Institute of Computer Science (SICS) since 1999, where he currently manages the network management and diagnostics group within the Industrial Applications and Methods (IAM) laboratory. His research interests are currently focused around network management, diagnostics, data mining and mobility modeling.
86
Communications IEEE
A
BEMaGS F
AVI MIRON (
[email protected]) _____________ is a researcher at the Computer Science Department of the Israeli Institute of Technology (Technion). Graduated from the University of Southern California, Los Angeles, he has participated in a few EU-funded research projects, including BIONETS and 4WARD, and now in SAIL and ETICS. He is an experienced high-tech executive and an entrepreneur in the area of tele/data communications, in both Israel and the United States. REBECCA STEINERT (
[email protected]) ________ is with IAM at SICS since 2006. She has a background in statistical machine learning and data mining, and in 2008 she received her M.Sc. from KTH in computer science with emphasis on autonomous systems. Since the beginning of 2010, she is pursuing her Ph.D. at KTH with focus on statistical approaches for network fault management. She has worked in the EU project 4WARD and is currently involved in SAIL, focusing on fault management in cloud computing and network virtualization. She also contributes to the network management research within the SICS Center for Networked Systems.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Palexpo, Geneva Exhibition Conference
19 - 21 September 18 - 22 September
Much more than a Conference and Exhibition Visitor Registration Now Open Register for FREE at www.ecocexhibition.com
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
TOPICS IN NETWORK AND SERVICE MANAGEMENT
Network Resilience: A Systematic Approach Paul Smith, Lancaster University David Hutchison, Lancaster University James P. G. Sterbenz, University of Kansas and Lancaster University Marcus Schöller, NEC Laboratories Europe Ali Fessi, Technische Universität München Merkouris Karaliopoulos, NKU Athens Chidung Lac, France Telecom (Orange Labs) Bernhard Plattner, ETH Zurich
ABSTRACT The cost of failures within communication networks is significant and will only increase as their reach further extends into the way our society functions. Some aspects of network resilience, such as the application of fault-tolerant systems techniques to optical switching, have been studied and applied to great effect. However, networks — and the Internet in particular — are still vulnerable to malicious attacks, human mistakes such as misconfigurations, and a range of environmental challenges. We argue that this is, in part, due to a lack of a holistic view of the resilience problem, leading to inappropriate and difficult-to-manage solutions. In this article, we present a systematic approach to building resilient networked systems. We first study fundamental elements at the framework level such as metrics, policies, and information sensing mechanisms. Their understanding drives the design of a distributed multilevel architecture that lets the network defend itself against, detect, and dynamically respond to challenges. We then use a concrete case study to show how the framework and mechanisms we have developed can be applied to enhance resilience.
INTRODUCTION Data communication networks are serving all kinds of human activities. Whether used for professional or leisure purposes, for safety-critical applications or e-commerce, the Internet in particular has become an integral part of our everyday lives, affecting the way societies operate. However, the Internet was not intended to serve all these roles and, as such, is vulnerable to a wide range of challenges. Malicious attacks, software and hardware faults, human mistakes (e.g., software and hardware misconfigurations), and
88
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
large-scale natural disasters threaten its normal operation. Resilience, the ability of a network to defend against and maintain an acceptable level of service in the presence of such challenges, is viewed today, more than ever before, as a major requirement and design objective. These concerns are reflected in, among other ways, in the Cyber Storm III exercise carried out in the United States in September 2010, and the “cyber stress tests” conducted in Europe by the European Network and Information Security Agency (ENISA) in November 2010; both aimed precisely at assessing the resilience of the Internet, this “critical infrastructure used by citizens, governments, and businesses.” Resilience evidently cuts through several thematic areas, such as information and network security, fault tolerance, software dependability, and network survivability. A significant body of research has been carried out around these themes, typically focusing on specific mechanisms for resilience and subsets of the challenge space. We refer the reader to Sterbenz et al. [1] for a discussion on the relation of various resilience disciplines, and to a survey by Cholda et al. [2] on research work for network resilience. A shortcoming of existing research and deployed systems is the lack of a systematic view of the resilience problem, that is, a view of how to engineer networks that are resilient to challenges that transcend those considered by a single thematic area. A non-systematic approach to understanding resilience targets and challenges (e.g., one that does not cover thematic areas) leads to an impoverished view of resilience objectives, potentially resulting in ill suited solutions. Additionally, a patchwork of resilience mechanisms that are incoherently devised and deployed can result in undesirable behavior and increased management complexity under chal-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
BEMaGS F
Challenges
Resilience objectives described, e.g., in SLAs
Used to determine whether targets are being met 4
1 Resilience target
Management decisions to control resilience mechanisms 5
Resilience estimator
Resilience manager
Networks and services Provided service
Resilience mechanisms
Detects and characterises challenges
5
3 Redundant and diverse infrastructure provisioning and self-protecting services
A
Challenge analysis
Protocols and services embedded in the network
Defensive measures 2
Figure 1. The resilience control loop: derived from the real-time component of the D2R2 + DR resilience strategy.
lenge conditions, encumbering the overall network management task [3]. The EU-funded ResumeNet project argues for resilience as a critical and integral property of networks. It advances the state of the art by adopting a systematic approach to resilience, which takes into account the wide-variety of challenges that may occur. At the core of our approach is a coherent resilience framework, which includes implementation guidelines, processes, and toolsets that can be used to underpin the design of resilience mechanisms at various levels in the network. In this article, we first describe our framework, which forms the basis of a systematic approach to resilience. Central to the framework is a control loop, which defines necessary conceptual components to ensure network resilience. The other elements — a risk assessment process, metrics definitions, policybased network management, and information sensing mechanisms — emerge from the control loop as necessary elements to realize our systematic approach. We show how these elements drive the design of a novel architecture and mechanisms for resilience. Finally, we illustrate these mechanisms in a concrete case study being explored in ResumeNet: a future Internet smart environments application.
FRAMEWORK FOR RESILIENCE Our resilience framework builds on work by Sterbenz et al. [1], whereby a number of resilience principles are defined, including a resilience strategy, called D 2 R 2 + DR: Defend, Detect, Remediate, Recover, and Diagnose and Refine. The strategy describes a real-time control loop to allow dynamic adaptation of networks in response to challenges, and a non-real time control loop that aims to improve the design of the network, including the real-time loop operation, reflecting on past operational experience. The framework represents our systematic approach to the engineering of network
resilience. At its core is a control loop comprising a number of conceptual components that realize the real-time aspect of the D2R2 + DR strategy, and consequently implement network resilience. Based on the resilience control loop, other necessary elements of our framework are derived, namely resilience metrics, understanding challenges and risks, a distributed information store, and policy-based management. The remainder of this section describes the resilience control loop, then motivates the need for these framework elements.
RESILIENCE CONTROL LOOP Based on the real-time component of the D2R2 + DR strategy, we have developed a resilience control loop, depicted in Fig. 1, in which a controller modulates the input to a system under control in order to steer the system and its output towards a desired reference value. The control loop forms the basis of our systematic approach to network resilience — it defines necessary components for network resilience from which the elements of our framework, discussed in this section, are derived. Its operation can be described using the following list; items correspond to the numbers shown in Fig. 1: 1. The reference value we aim to achieve is expressed in terms of a resilience target, which is described using resilience metrics. The resilience target reflects the requirements of end users, network operators, and service providers. 2. Defensive measures need to be put in place proactively to alleviate the impact of challenges on the network, and maintain its ability to realize the resilience target. A process for identifying the challenges that should be considered in this defense step of the strategy (e.g., those happening more frequently and having high impact) is necessary. 3. Despite the defensive measures, some challenges may cause the service delivered to users to deviate from the resilience target. These challenges could include unforeseen attacks or mis-
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
89
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Severely degraded
Impaired
Sc
Remediate
Sr
Detect Acceptable
Service parameters P
Unacceptable
Partially degraded
Defend
Recover
S0
Figure 2. Resilience state space
configurations. Challenge analysis components detect and characterize them using a variety of information sources. 4. Based on output from challenge analysis and the state of the network, a resilience estimator determines whether the resilience target is being met. This measure is based on resilience metrics, and is influenced by the effectiveness of defense and remediation mechanisms to respond to challenges. 5. Output from the resilience estimator and challenge analysis is fed to a resilience manager. It is then its responsibility to control resilience mechanisms embedded in the network and service infrastructure, to preserve the target service provision level or ensure its graceful degradation. This adaptation is directed using resilience knowledge, not shown in Fig. 1, such as policies and challenge models. We anticipate a cost of remediation in terms of a potentially unavoidable degradation in quality of service (QoS), which should not be incurred if the challenge abates. Consequently, the network should aim to recover to normal operation after a challenge has ceased. The purpose of the background loop in the D2R2 + DR strategy is to improve the operation of the resilience control loop such that it meets an idealized system operation. This improvement could be in response to market forces, leading to new resilience targets, new challenges, or suboptimal performance. The diagnose phase identifies areas for improvement, including defense, that are enacted through refinement. In reality, and for the foreseeable future, we anticipate this outer loop to be realized with human intervention.
90
Communications IEEE
BEMaGS F
RESILIENCE METRICS
Operational state N Normal operation
A
Defining a resilience target requires appropriate metrics. Ideally, we would like to express the resilience of a network using a single value, R, in the interval [0,1], but this is not a simple problem because of the number of parameters that contribute to and measure resilience, and due to the multilayer aspects in which each level of resilience (e.g., resilient topology) is the foundation for the next level up (e.g., resilient routing). We model resilience as a two-dimensional state space in which the vertical axis P is a measure of the service provided when the operational state N is challenged, as shown in Fig. 2. Resilience is then modeled as the trajectory through the state as the network goes from delivering acceptable service under normal operations S 0 to degraded service S c . Remediation improves service to S r and recovery returns to the normal state S0. We can measure resilience at a particular service level as the area under this trajectory, R. We have developed a number of tools for evaluating network resilience. For example, we use MATLAB or ns-3 simulation models to measure the service at each level and plot the results under various challenges and attacks, as in Fig. 2, where each axis is an objective function of the relevant parameters [4]. Furthermore, we have developed the Graph Explorer tool [5] that takes as input a network topology and associated traffic matrix, a description of challenges, and a set of metrics to be evaluated. The result of the analysis is a series of plots that show the metric envelope values (m i (min), m i (max)) for each specified metric mi, and topology maps indicating the resilience across network regions. Figure 3 shows an example of the resilience of the European academic network GÉANT2 to link failures. The set of plots in Fig. 3a show metric envelopes at different protocol levels — the aim is to understand how jitter responds in comparison with metrics at other levels, such as queue length and connectivity. Surprisingly, jitter is not clearly related to queue length, and a monotonic increase in path length does not yield a similar increase in queue length for all scenarios of link failures. In fact, the fourth link failure disconnects a region of the network; whereas up to three failures, the heavy use of a certain path resulted in increasing queue lengths and jitter. The partition increases path length, because route lengths are set to infinity, and decreases connectivity, which is accompanied by a reduction in jitter, shown with the blue arrows in Fig. 3a. The topology map in Fig. 3b highlights the vulnerability of regions of GÉANT2 with a heat map, which can be used by network planners. Our framework for resilience metrics (i.e., the multilevel two-dimensional state space and the use of metric envelopes) can be used to understand the resilience of networks to a broad range of challenges, such as misconfigurations, faults, and attacks. The ability to evaluate a given network’s resilience to a specific challenge is limited by the capability of the tools to create complex challenge scenarios — this is an area for further work, in which our effort should be focused on modeling pertinent high-impact challenges.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
F
0.01
0
1
2 Failures
3
4
1
2 Failures
3
4
Level 1 physical
Connectivity
Level 2 link 1 0.99 0.98 0.97 0.96 0.95 0.94 0.93
0
1
2 Failures
3
80 75 70 65 60 55 50
0
1
2 Failures
3
0.28 0.26 0.24 0.22 0.2 0.18 0.16 0.13
1
2 Failures
3
4
2 Failures
3
4
4
0 Clustering
Path length 0
Betweenness
11 9 7 5 3
Connectivity Cluttering coefficient
Level 3 routing
Hop count
14
0
Betweenness
Level 4 transport
1400 1200 1000 800 600 400 200 0
Queue length
0.011
0
1
2 Failures
3
4
-0.05 -0.1
Assortativity
0.012
Queue length
Level 7 application
0.013
End-to-end jitter
Jitter
0.014
-0.15 -0.2 -0.27
0
1
4 Network design Network operation (a)
(b)
Figure 3. Example output from the Graph Explorer, developed in the ResumeNet project: a) plots showing the relationship between metrics at various layers in response to link failures on the GÉANT2 topology; b) a heat map showing vulnerable regions of the topology with respect to a given set of metrics. Reprinted from [5].
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
91
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
UNDERSTANDING CHALLENGES AND RISKS We advocate the use of a policy-based management framework to define the behavior of real-time loop instantiations. Consequently, the implementation of resilience mechanisms can be decoupled from the resilience management strategies, which are expressed in policies.
Engineering resilience has a monetary cost. To maximize the effectiveness of the resources committed to resilience, a good understanding of the challenges a network may face is mandatory. We have developed a structured risk assessment approach that identifies and ranks challenges in line with their probability of occurrence and their impact on network operation (i.e., how disruptive they are to the provision of its services). The approach should be carried out at the stage of network design when proactive defensive measures are deployed, and repeated regularly over time as part of the process of network improvements. Central to determining the impact of a challenge is to identify the critical services the network provides and the cost of their disruption: a measure of impact. Various approaches can be used to identify the critical services, such as discussion groups involving the network’s stakeholders. Networked systems are implemented via a set of dependent subsystems and services (e.g., web and Session Initiation Protocol [SIP] services rely on Domain Name Service [DNS]). To identify whether challenges will cause a degradation of a service, it is necessary to explicate these dependencies. The next phase is to identify the occurrence probabilities of challenges (challenge_prob). Some challenges will be unique to a network’s context (e.g., because of the services it provides), while others will not. In relation to these challenges, shortcomings of the system (e.g., in terms of faults) should be identified. The aim is to determine the probability that a challenge will lead to a failure (fail_prob). We can use tools, such as our Graph Explorer, analytical modeling, and previous experience (e.g., in advisories) to help identify these probabilities. Given this information, a measure of exposure can be derived using the following equation: exposure = (challenge_prob × fail_prob) × impact With the measures of exposure at hand, resilience resources can be targeted at the challenges that are likely to have the highest impact.
INFORMATION SOURCES AND SHARING FOR RESILIENCE For the most part, network management decisions are made based on information obtained from monitoring systems in the network (e.g., via Simple Network Management Protocol [SNMP]). However, to be able to make autonomic decisions about the nature of a wide range of challenges and how to respond to them — a necessary property of resilient networks — a broader range of information needs to be used. In addition to traditional network monitoring information, context information, which is sometimes “external” to the system can be used. Earlier work has demonstrated how the use of weather information, an example of context, improves the resilience of millimeter-wave wireless mesh networks, which perform poorly in heavy rain [4]. Also, in addition to node-centric monitoring tools, such as NetFlow and SNMP, task-centric tools can be used to determine the
92
Communications IEEE
A
BEMaGS F
root cause of failures. For example, X-trace [6] is a promising task-centric monitoring approach that can be used to associate network and service state (e.g., router queue lengths and DNS records) with service requests (e.g., retrieving a web page). This multilevel information can then be used to determine the root causes of failures. We are developing a Distributed Store for Challenges and their Outcome (DISco), which uses a publish-subscribe messaging pattern to disseminate information between subsystems that realize the real-time loop. Such information includes actions performed to detect and remediate challenges. Information sources may report more data than we can afford or wish to relay on the network, particularly during challenge occurrences. DISco is able to aggregate information from multiple sources to tackle this problem. Decoupling information sources from components that use them allows adaptation of challenge analysis components without needing to modify information sources. To assist the two phases of the outer loop, DISco employs a distributed peer-to-peer storage system for longerterm persistence of data, which is aware of available storage capacity and demand.
POLICIES FOR RESILIENCE We advocate the use of a policy-based management framework to define the behavior of realtime loop instantiations. Consequently, the implementation of resilience mechanisms can be decoupled from the resilience management strategies, which are expressed in policies. This has two immediate benefits: the nature of challenges changes over time — management strategies can be adapted accordingly without the need for network down-time; and policies allow network operators to clearly express when they would like to intervene in the network’s operation (e.g., when a remediation action needs to be invoked). Research outcomes from the policy-based management field can help address the complexities of resilience management [7]. A difficult task is deriving implementable policies from high-level resilience requirements, say, expressed in service level agreements (SLAs). With appropriate modifications, techniques for policy refinement can be used to build tools to automate aspects of this process. Policy-based learning, which relies on the use of logical rules for knowledge representation and reasoning, is being exploited to assist with the improvement stages of our strategy. Techniques for policy ratification are currently used to ensure that invocation of different resilience strategy sets does not yield undesirable conflicting behavior. Conflicts can occur horizontally between components that realize the resilience control loop, and vertically across protocol levels. For example, a mechanism that replicates a service using virtualization techniques at the service level could conflict with a mechanism that is rate-limiting traffic at the network level. Example policies of this sort are shown in Fig. 4. So that these forms of conflict can be detected, Agrawal et al. [8] provide a theoretical foundation for conflict resolution that needs to be extended with domain-specific knowledge, for example, regarding the nature of resilience mechanisms.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
F
Since challenges may
on highUtilisation (link) { do { RateLimiterMO limit (link , 50%) } }
vary broadly from topology-level link Conflict ServerB
Virtualised services
failures to application-level malware, defensive measures against anticipated high-impact
ServerA
challenges need to be applied at
Internet
different levels and locations.
DDoS attackers DDoS traffic Replication traffic
on highServiceUtilisation (service) { do { VMReplicator replicateService (service, ServerB) } }
Figure 4. Potentially conflicting policies at the service level (the replication of a service) and the network level (rate-limiting traffic) that could be triggered by the same challenge, such as a distributed denial of service (DDoS) attack. Rate limiting traffic could cause the replication to fail.
DEFENSE AND DYNAMIC ADAPTATION ARCHITECTURE In this section, we describe a set of defensive mechanisms and an architecture that realize our systematic approach to resilience, described earlier. The architecture, shown in Fig. 5, consists of several subsystems implementing the various tasks of the communication system as well as the challenge detection components and adaptation capabilities. The behavior of all these subsystems is directed by the resilience manager using policies, which are held in a resilience knowledge base. Central to this architecture is DISco, which acts as a publish-subscribe and persistent storage system, containing information regarding ongoing detection and remediation activities. From an implementation perspective, based on the deployment context, we envisage components of the architecture to be distributed (e.g., in an Internet service provider [ISP] network) or functioning entirely on a single device (e.g., nodes in a delay-tolerant network).
DEFENSIVE MEASURES As a first step, defensive measures need to be put in place to alleviate the impact of challenges on the network. Since challenges may vary broadly from topology-level link failures to application-level malware, defensive measures against anticipated high-impact challenges need to be applied at different levels and locations: in the network topology design phase, and within protocols; across a network domain, as well as at individual nodes. Defensive measures can either prevent a challenge from affecting the system or contain erroneous behavior within a subsystem in such a way that the delivered service still
meets its specification. A selection of defensive measures developed in the ResumeNet project is shown in Table 1.
DETECTION SUBSYSTEMS The second step is to detect challenges affecting the system leading to a deviation in delivered service. We propose an incremental approach to challenge analysis. Thereby, the understanding about the nature of a challenge evolves as more inputs become available from a variety of information sources. There are two apparent advantages of this incremental approach. First, it readily accommodates the varying computational overhead, timescales, and potentially limited accuracy of current detection approaches [9]. Second, relatively lightweight detection mechanisms that are always on can be used to promptly initiate remediation, thus providing the network with a first level of protection, while further mechanisms are invoked to better understand the challenge and improve the network response. Lightweight detection mechanisms can be driven by local measurements carried out in the immediate neighborhood of affected nodes. For example, consider high-traffic volume challenges, such as a DDoS attack or a flash crowd event. Initially, always-on simple queue monitoring could generate an alarm if queue lengths exceed a threshold for a sustained period. This could trigger the rate limiting of links associated with high traffic volumes. More expensive traffic flow classification could then be used to identify and block malicious flows, consequently not subjecting benign flows to rate limiting. Challenge models, shown in Fig. 5, describe symptoms of challenges and drive the analysis process. They can be used to initially identify broad classes of challenge, and later to refine identification to more specific instances.
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
93
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
generic approach to resilience through concrete study cases
Resilience knowledge base Inform Challenge models
Policies
Inform
paradigms: wireless
peer-to-peer voice
tiv
Subscribe(challenge) / publish(report) / lookup(alarm)
Challenge analysis
Consultants
Respond
Deliver (challenge)R e
at
e
gis
ter
Subscribe(alarm) / publish(challenge)
mesh and delay tolerant networks,
F
Request
Network resilience manager
Ac
that cover a range of future networking
BEMaGS
Remediation and recovery
We are currently evaluating our
A
Managed entities
DISco Deliver(alarm)
conferencing and
Publish(alarm)
service provision over heterogeneous smart
Apply
e
at pd
U Information sources
environments.
Detection
Defensive measures
Defend
Figure 5. A dynamic adaptation architecture that realizes the resilience control loop.
REMEDIATION AND RECOVERY SUBSYSTEMS The challenge detection subsystem interfaces with the remediation and recovery subsystem, the third and final step, by issuing alerts to DISco using the publish(challenge) primitive. These alerts contain information about the challenge and its impact on the network, in terms of the metrics that are falling short of the resilience target. The network resilience manager takes this information as context data, and, based on policies, selects an adaptation strategy. In doing so, the network resilience manager realizes the resilience management functionality in Fig. 1. If further information is required by the network resilience manager that is not contained in the alert, the lookup(alarm) primitive can be used. Furthermore, the network resilience manager can make use of consultants, such as path computation elements, which can compute new topological configurations, such as new channel allocations or new forwarding structures. Resilience mechanisms are deployed by enforcing new configurations on the managed entities (e.g., routers and end hosts) in the network. To implement the resilience estimator, the network resilience manager assesses the success of chosen remedies. The assessment is stored in DISco to aid the diagnosis and refinement steps of the background loop. Carrying out this assessment is not straightforward since it requires spatio-temporal correlation of changes in network state, which is an issue for further work.
RESILIENCE IN SMART ENVIRONMENTS: A CASE STUDY We are currently evaluating our generic approach to resilience through concrete study cases that cover a range of future networking
94
Communications IEEE
paradigms: wireless mesh and delay-tolerant networks, peer-to-peer voice conferencing, and service provision over heterogeneous smart environments. Herein, we focus our discussion on the last study case. The widespread use of smart mobile devices, together with identifiers such as radio frequency identification (RFID), embedded in objects such as products, enables communication with, and about, these objects. The French national project Infrastructure for the Future Trade (ICOM) has developed an intra- and interenterprise infrastructure, depicted in Fig. 6, that allows the connection of objects with enterprise information systems and fixed or mobile terminals. This ICOM platform can be used as a foundation for a number of enterprise applications. The experimentation makes use of three different entities: • The data acquisition site is the data source — items identified by RFID, for example, are read and their information sent to a processing centre located remotely. • The data processing site houses different modules of the platform (e.g., data collection, aggregation, and tracking), which will forward the enriched data to the core application. • The application provision site hosts the platform’s central element — it is also where the data subscriber applications (web services, legal application, etc.) are linked. Based on outcomes of our risk assessment approach, high-impact challenges to the platform include those that are intentional and accidental: malicious attacks that threaten the confidentiality and integrity of commercially sensitive data, DDoS attacks by extortionists, and, given the immature nature of the platform, software and hardware faults. This understanding ensures that we implement appropriate
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
F
Defensive measure
Description
Innovation
Survivable Network Design (SND) [11]
During network planning, SND optimizes network operations, such as routing and transport, in the presence of high-impact challenges.
Expansion of the methodology to derive a cooperationfriendly routing scheme for Wireless Mesh Networks (WMNs) to cope with node selfishness, explicitly accounting for radio interference constraints [12].
Game-theoretical node protection [13]
Node protection schemes are deployed against propagation of malware, which may compromise network nodes and threaten the network resilience.
The game-theoretic formulation of the problem confirms heavy dependence on the underlying topology and allows for optimal tuning of node protection level.
Rope-ladder routing [14]
Multi-path forwarding structure combining link and node protection in a way that the loss gap and QoS penalty, e.g., delay, during fail-over is minimized.
Better use of path diversity for support of real-time traffic, e.g., voice flows, for which burst packet loss during the path recovery time matters.
Cooperative SIP (CoSIP) [15]
An extension of the Session Initiation Protocol (SIP), whereby endpoints are organized into a peer-topeer (P2P) network. The P2P network stores location information and is used when the SIP server infrastructure is unavailable.
Optimal setting of the number of replica nodes in the P2P network for given service reliability levels, inline with an enhanced trace-driven reliability model.
Virtual service migration
Enables redundancy and spatial diversity by relocating service instances on-the-fly, such that a continuous acceptable service can be provided to its users.
Existing approaches are tailored toward resilience to hardware failures within data centers. The derivation of service migration strategies from migration primitives, providing resilience against a variety of challenges.
Table 1. A selection of defensive measures developed in the ResumeNet Project.
defensive measures and dynamic adaptation strategies. Consequently, defensive measures primarily include secure VPN connections between sites, enabling confidentiality and integrity of the data in transit. Security mechanisms, such as authentication and firewalls, are also implemented. Redundancy of infrastructure and implementation diversity of services are exploited to maintain reliability and availability in the presence of failures caused by software faults. Incremental challenge analysis is realized using the Chronicle Recognition System (CRS), a temporal reasoning system aimed at alarmdriven automated supervision of data networks [10]. Lightweight detection mechanisms generate alarms based on metrics, such as anomalous application response times and data processing request rates. Finally, policy-based adaptation, implementing remediation and recovery, is achieved through the specification of the platform’s nominal and challenge context behavior (i.e., its configuration in response to anticipated challenges). In our case study, challenge context policies describe configurations in response to alarms indicating a DDoS attack. For example, modified firewall configurations are defined to block traffic deemed to be malicious; service virtualization configurations that make use of redundant infrastructure are also specified to load balance increased resource demands. The transition between behaviors is based on alert messages, generated via challenge analysis, and outcomes from continuous threat level assessment. The case study sketched above illustrates the gain from applying our resilience strategies in a systematic approach: starting from a risk assessment, challenges are derived, allowing defense measures to be deployed. The following step is the specification of chronicles —
temporal descriptions of challenges — for detection by the CRS, and policy-driven mechanisms to remediate and recover from unforeseen failures.
CONCLUSION Given the dependence of our society on network infrastructures, and the Internet in particular, we take the position that resilience should be an integral property of future networks. In this article, we have described a systematic approach to network resilience. Aspects of our work represent a longer-term vision of resilience and necessitate more radical changes in the way network operators currently think about resilience. Further experimentation and closer engagement with operators through initiatives like ENISA, which focus on the resilience of public communication networks and services, are required before some of this research becomes standard practice. On the other hand, application-level measures, such as service virtualization, necessitate fewer changes at the network core and lend to easier implementation. Further benefits for network practitioners are anticipated through the use of tools like the Graph Explorer, which can explore correlations among metrics at various levels of network operation.
ACKNOWLEDGEMENTS The work presented in this article is supported by the European Commission under Grant No. FP7-224619 (the ResumeNet project). The authors are grateful to the members of the ResumeNet consortium, whose research has contributed to this article, and in particular to Christian Doerr and his colleagues at TU Delft for the work presented in Fig. 3.
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
95
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Data acquisition site
A
BEMaGS F
Acquisition
Application provision site
1D barcode
Scanner
NFC mobile
Payment device
Legal applications
Defend
Mobile photo
Defend
2D barcode
Web applications Web services Software packages
VPN
Defend
RFID reader
VPN
RFID tag
Data processing site
VPN
Transport
Defend
Traceability
Defend
Collection aggregation
Internet Remediate and recover Alarm Nominal behaviour
Persistant storage
Challenge contexts Timeout
Figure 6. The ICOM platform connecting enterprise sites that perform data processing and application provisioning with objects in a smart environment. Selected resilience mechanisms are shown that can be used to mitigate identified challenges.
REFERENCES [1] J. P. G. Sterbenz et al., “Resilience and Survivability in Communication Networks: Strategies, Principles, and Survey of Disciplines,” Elsevier Computer Networks, Special Issue on Resilient and Survivable Networks, vol. 54, no. 8, June 2010, pp. 1243–42. [2] P. Cholda et al., “A Survey of Resilience Differentiation Frameworks in Communication Networks,” IEEE Commun. Surveys & Tutorials, vol. 9, no. 4, 2007, pp. 32–55. [3] ENISA Virtual Working Group on Network Providers’ Resilience Measures, “Network Resilience and Security: Challenges and Measures,” tech. rep. v1.0, Dec. 2009. [4] J. P. G. Sterbenz et al., “Evaluation of Network Resilience, Survivability, and Disruption Tolerance: Analysis, Topology Generation, Simulation, and Experimentation (invited paper),” Springer Telecommun. Sys., 2011, accepted Mar. 2011. [5] C. Doerr and J. Martin-Hernandez, “A Computational Approach to Multi-Level Analysis of Network Resilience,” Proc. 3rd Int’l. Conf. Dependability, Venice, Italy, July 2010. [6] R. Fonseca et al., “X-trace: A Pervasive Network Tracing Framework,” 4th USENIX Symp. Networked Sys. Design & Implementation, Santa Clara, CA, June 2007, pp. 271–84.
96
Communications IEEE
[7] P. Smith et al., “Strategies for Network Resilience: Capitalizing on Policies,” AIMS 2010, Zürich, Switzerland, June 2010, pp. 118–22. [8] D. Agrawal et al., “Policy Ratification,” 6th IEEE Int’l. Wksp. Policies for Distrib. Sys. and Networks, Stockholm, Sweden, June 2005, pp. 223–32. [9] V. Chandola, A. Banerjee, and V. Kumar, “Anomaly Detection: A Survey,” ACM Comp. Surveys, vol. 41, July 2009, pp. 1–58. [10] M.-O. Cordier and C. Dousson, “Alarm Driven Monitoring Based on Chronicles,” 4th Symp. Fault Detection, Supervision and Safety for Technical Processes, Budapest, Hungary, June 2000, pp. 286–91. [11] E. Gourdin, “A Mixed-Integer Model for the Sparsest Cut Problem,” Int’l. Symp. Combinatorial Optimization, Hammamet, Tunisia, Mar. 2010, pp. 111–18. [12] G. Popa et al., “On Maximizing Collaboration in Wireless Mesh Networks Without Monetary Incentives,” 8th Int’l. Symp. Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks, May 2010, pp. 402–11. [13] J. Omic, A. Orda, and P. Van Mieghem, “Protecting Against Network Infections: A Game Theoretic Perspective,” Proc. 28th IEEE INFOCOM, Rio de Janeiro, Brazil, Apr. 2009, pp. 1485–93. [14] J. Lessman et al., “Rope Ladder Routing: PositionBased Multipath Routing for Wireless Mesh Networks,” Proc. 2nd IEEE WoWMoM Wksp. Hot Topics in Mesh Networking, Montreal, Canada, June 2010, pp. 1–6.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
[15] A. Fessi et al., “A Cooperative SIP Infrastructure for Highly Reliable Telecommunication Services,” ACM Conf. Principles, Sys. and Apps. of IP Telecommun., New York, NY, July 2007, pp. 29–38.
BIOGRAPHIES PAUL SMITH is a senior research associate at Lancaster University’s School of Computing and Communications. He submitted his Ph.D. thesis in the area of programmable networking resource discovery in September 2003, and graduated in 1999 with an honors degree in computer science from Lancaster. In general, he is interested in the various ways that networked (socio-technical) systems fail to provide a desired service when under duress from various challenges, such as attacks and misconfigurations, and developing approaches to improving their resilience. In particular, his work has focused on the rich set of challenges that face community-driven wireless mesh networks. DAVID HUTCHISON is director of InfoLab21 and professor of computing at Lancaster University, and has worked in the areas of computer communications and networking for more than 25 years, recently focusing his research efforts on network resilience. He has served as member or chair of numerous TPCs (including the flagship ACM SIGCOMM and IEEE INFOCOM), and is an editor of the renowned Springer Lecture Notes in Computer Science and the Wiley CNDS book series. J AMES P. G. S TERBENZ is director of the ResiliNets research group at the Information & Telecommunication Technology Center and associate professor of electrical engineering and computer science at The University of Kansas, a visiting professor of computing in InfoLab21 at Lancaster University, and has held senior staff and research management positions at BBN Technologies, GTE Laboratories, and IBM Research. He received a doctorate in computer science from Washington University in St. Louis, Missouri. His research is centered on resilient, survivable, and disruptiontolerant networking for the future Internet for which he is involved in the NSF FIND and GENI programs as well as the EU FIRE program. He is principal author of the book HighSpeed Networking: A Systematic Approach to High-Bandwidth Low-Latency Communication. He is a member of the ACM, IET/IEE, and IEICE. MARCUS SCHÖLLER is a research scientist at NEC Laboratories Europe, Germany. He received a diploma in computer science from the University of Karlsruhe, Germany, in 2001 and his doctorate in engineering in 2006 on robustness and stability of programmable networks. Afterward he
IEEE
BEMaGS F
held a postdoc position at Lancaster University, United Kingdom, focusing his research on autonomic networks and network resilience. He is currently working on resilience for future networks, fault management in femtocell deployments, and infrastructure service virtualization. His interests also include network and system security, intrusion detection, self-organization of networks, future network architectures, and mobile networks including mesh and opportunistic networks. A LI F ESSI is a researcher at the Technische Universität München (TUM). He holds a Ph.D. from TUM and a Diplom (Master’s) from the Technische Universität Kaiserslautern. His research currently focuses on the resilience of network services, such as web and SIP, using different techniques (e.g., P2P networking, virtualization, and cryptographic protocols). He is a regular reviewer of several scientific conferences and journals, such as ACM IPTComm, IEEE GLOBECOM, IFIP Networking, and IEEE/ACM Transactions on Networking. M ERKOURIS K ARALIOPOULOS is a Marie Curie Fellow in the Department of Informatics and Telecommunications, National and Kapodistrian University of Athens, Greece, since September 2010. He was a postdoctoral researcher in the University of North Carolina, Chapel Hill, in 2006 and a senior researcher and lecturer at ETH Zurich, Switzerland, from 2007 until 2010. His research interests lie in the general area of wireless networking, currently focusing on network resilience problems related to node selfishness and misbehavior. C HIDUNG L AC is a senior researcher at France Telecom (Orange Labs). Besides activities linked with network architecture evolution, for which he contributes to the design of scenarios and roadmaps, his research interests are centered on network and services resilience, particularly through his involvement in European projects such as the ReSIST Network of Excellence (2006–2009) and the present STREP ResumeNet (2008–2011). He holds a Doctorat d’Etat-és-Sciences Physiques (1987) from the University of Paris XI Orsay. BERNHARD PLATTNER is a professor of computer engineering at ETH Zurich, where he leads the communication systems research group. He has a diploma in electrical engineering and a doctoral degree in computer science from ETH Zurich. His research currently focuses on self-organizing networks, systems-oriented aspects of information security, and future Internet research. He is the author or co-author of several books and has published over 160 refereed papers in international journals and conferences.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
97
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
TOPICS IN NETWORK AND SERVICE MANAGEMENT
A Survey of Virtual LAN Usage in Campus Networks Minlan Yu and Jennifer Rexford, Princeton University Xin Sun and Sanjay Rao, Purdue University Nick Feamster, Georgia Institute of Technology
ABSTRACT VLANs are widely used in today’s enterprise networks to improve Ethernet scalability and support network policies. However, manuals and textbooks offer very little information about how VLANs are actually used in practice. Through discussions with network administrators and analysis of configuration data, we describe how three university campuses and one academic department use VLANs to achieve a variety of goals. We argue that VLANs are ill-suited to some of these goals (e.g., VLANs are often used to realize access control policies, but constrain the types of policies that can be expressed). Furthermore, the use of VLANs leads to significant complexity in the configuration of network devices.
INTRODUCTION Enterprise networks, which connect the computers within a college campus or corporate location, differ markedly from backbone networks. These networks have distinctive topologies, protocols, policies, and configuration practices. Yet, the unique challenges in enterprise networks are not well understood outside of the operator community. One prominent example is virtual LANs (VLANs) — a widely- used technology that is barely discussed in networking textbooks. VLANs were initially intended to allow network administrators to connect a group of hosts in the same broadcast domain, independent of their physical location. However, today’s enterprise administrators use VLANs for a variety of other purposes, most notably for better scalability and flexible specification of policies. However, enterprise administrators have seen many problems of VLANs because VLANs are used for other functions they were not designed for. Understandably, VLANs are at best an incomplete solution for some of these problems. As a result, managing VLANs is one of the most challenging tasks they face. In this article, we study four networks — three university campuses and one academic department — to better understand how VLANs are used in practice. Through discussions with network administrators, and targeted analysis of
98
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
router configuration data, we have obtained deeper insights into how the administrators use VLANs to achieve a variety of design goals, and the difficulties they encounter in the process. We show that VLANs are not well-suited for many of the tasks that they support today, and argue that future enterprise network architectures should decouple policy specification from scalability concerns with layer-2 protocols, topology, and addressing. After a brief survey of VLAN technology, we describe how the four networks use VLANs to support resource isolation, access control, decentralized management, and host mobility. However, VLANs were not designed with these goals in mind — network administrators use VLANs for the lack of a better alternative. We argue that VLANs are too crude a mechanism for specifying policies, due to scalability constraints (on the number and size of VLANs) and the coarsegrained ways of assigning traffic to different VLANs. Further, VLAN configuration is far too complicated, due to the tight coupling with spanning-tree construction, failure recovery, host address assignment, and IP routing, as discussed. We conclude the article.
VIRTUAL LOCAL AREA NETWORKS An enterprise network consists of islands of Ethernet switches connected both to each other and to the rest of the Internet by IP routers, as shown in Fig. 1. We describe how administrators group related hosts into VLANs, and how the switches and routers forward traffic between hosts.
CONVENTIONAL LOCAL AREA NETWORKS In a traditional local area network (LAN), hosts are connected by a network of hubs and switches. The switches cooperate to construct a spanning tree for delivering traffic. Each switch forwards Ethernet frames based on its destination MAC address. If the switch contains no forwarding-table entry for the frame’s destination MAC address, the switch floods each frame over the entire spanning tree. A switch learns how to reach a MAC address by remembering the
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
incoming link for frames sent by that MAC address and creating a mapping between the MAC address and that port. To connect to the rest of the enterprise network (and the rest of the Internet), the island of Ethernet switches connects to IP routers that forward traffic to and from remote hosts. Each host interface in the LAN has an IP address from a common IP prefix (or set of prefixes). Traffic sent to an IP address in the same subnet stays within the LAN; the sending host uses the Address Resolution Protocol (ARP) to determine the MAC address associated with the destination IP address. For traffic destined to remote IP addresses, the host forwards the packets to the gateway router, which forwards packets further toward their destinations.
COMMUNICATION WITHIN A VLAN Administrators use VLANs to construct network segments that behave logically like a conventional LAN but are independent of the physical locations of the hosts; for example, hosts H1 and H3 in Fig. 1 both belong to VLAN1. As in a conventional physical LAN, the switches in a VLAN construct a spanning tree, and use flooding and learning to forward traffic between hosts. For example, the switches S3, S4, and S5 form a spanning tree for VLAN2. Communication between hosts in the same VLAN stays within the VLAN, with the switches forwarding Ethernet frames along the spanning tree to the destination MAC address. For example, hosts H2 and H4 communicate over the 2 spanning tree in VLAN2 based on their MAC addresses. Similarly, hosts H1 and H3 communicate over the spanning tree in VLAN1, where some of the IP routers (e.g., R1, R2, and R2) may also act as switches in the spanning tree; alternatively, a tunnel between R1 and R2 could participate in VLAN1 so the links in the IP backbone do not need to participate in the VLANs.
COMMUNICATION BBETWEEN VLANS Each host has an IP address from an IP prefix (or prefixes) associated with its VLAN; IP routers forward packets based on these prefixes, over paths computed in the routing protocol (e.g., Open Shortest Path First [OSPF] or Routing Information Protocol [RIP]). Hence, traffic between hosts in different VLANs must traverse an intermediate IP router. For example, traffic between hosts H3 and H4 would traverse router R2, even though the two hosts connect to the same switch. For example, when sending traffic to H4, host H3 forwards the packets to its gateway router R2, since the destination IP address belongs to a different prefix. R2 would then look up the destination IP address to forward the packet to H4 in VLAN2. If H4 sends an IP packet to H1, then H4’s router R3 forwards the packet based on the IP routing protocol toward the router announcing H1’s IP prefix, and that router would then forward the packet over the spanning tree for VLAN1.
CONFIGURING VLAN PORTS Supporting VLANs requires a way to associate switch ports with one or more VLANs. Administrators configure each port as either an access
IEEE
BEMaGS
Ethernet island
VLAN2
R2 S2
R1 R3
IP router backbone VLAN1
H1
VLAN1
S5 S4
S1 R5 Ethernet island
F
H3
H4
S6 S3
VLAN2 R4 Ethernet island
H2
Figure 1. Enterprise network with Ethernet islands interconnected by IP routers. port, which is connected to a host; or a trunk port, which is connected to another switch. An access port typically transports traffic for a single VLAN; the VLAN associated with a port may be either statically configured or dynamically assigned when the host connects, based on the host’s MAC address (e.g., using VLAN Management Policy Server VMPS [1]). In either case, the access port can tag incoming frames with the 12-bit VLAN identifier and removes the tag from outgoing frames, obviating the need for the hosts to support VLANs. In contrast, a trunk port may carry traffic for multiple VLANs; for example, switch S4’s port connecting to S5 must forward traffic for both VLAN1 and VLAN2 (and participate in each VLAN’s spanning tree protocol), but the trunk port to S3 does not. The administrators either manually configure each trunk port with a list of VLAN identifiers, or run a protocol like VLAN Trunking Protocol (VTP) [2] or Multiple VLAN Registration Protocol (MVRP) [3] to automatically determine which VLANs a trunk link should handle. Configuring a VLAN also requires configuring the gateway router to announce the associated IP prefixes into the routing protocol; each host interface must be assigned an IP address from the prefix associated with its VLAN.
VLAN USAGE IN CAMPUS NETWORKS Our campus network administrators use VLANs to achieve four main policy objectives — limiting the scope of broadcast traffic, simplifying access control policies, supporting decentralized network management, and enabling seamless host mobility for wireless users. The four networks include two large universities (campuses 1 and 2) and a department network (campus 3) within another university-wide network (campus 4). All four networks primarily run IPv4, with relatively limited experimental deployment of IPv6.
SCOPING BROADCAST TRAFFIC VLANs enable administrators to limit the scope of broadcast traffic and network-wide flooding, to reduce network overhead and enhance both privacy and security.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
99
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
VLANs provide an effective way to enforce access control by directing inter-VLAN traffic through routers. In addition, by allowing administrators to assign related hosts to IP addresses in the same subnet, VLANs simplify access control configuration by making packet-classification rules more concise.
Limiting the Broadcast/Flooding Overhead — End hosts broadcast Dynamic Host Configuration Protocol (DHCP) traffic when joining the LAN, and routinely broadcast Address Resolution Protocol (ARP) requests to learn the medium access control (MAC) addresses of other hosts in the same IP subnet. For example, campus 2 has one IP subnet with up to 4000 hosts with around 300 packets/s of broadcast traffic; this broadcast traffic is dominated by ARP, iTunes broadcast messages, and NetBios. It not only consumes network bandwidth, but also consumes bandwidth and energy resources on the end hosts (particularly for mobile devices). Switches also flood packets to a destination MAC address they have not yet learned how to reach. This consumes bandwidth resources, especially if the switches’ forwarding tables are not large enough to store an entry for each MAC address on the LAN. Administrators often divide large networks into multiple VLANs to limit the scope of broadcast messages and flooding traffic. For example, campuses 1 and 4 assign each building a different IP subnet, each associated with its own VLAN. The resulting broadcast domains are small enough to limit the overhead on the switches and the end hosts. Protecting Security and Privacy — Broadcast and flooding traffic also raise security and privacy concerns. Sending excessive broadcast traffic is an effective denial-of-service attack on the network. In addition, a malicious host can intentionally overload switch forwarding tables (e.g., by spoofing many source MAC addresses), forcing switches to flood legitimate traffic that can be easily monitored by the attacking host. ARP is also vulnerable to man-in-the-middle attacks, where a malicious host sends unsolicited ARP responses to impersonate another host on the LAN, thereby intercepting all traffic sent to the victim. Network administrators can reduce these risks by constraining which users can belong to the same VLAN. For example, campus 3 has separate subnets for faculty, graduate students, and undergraduate students, and assigns each subnet to one VLAN based on the registered MAC addresses of the user machines. This ensures that students cannot intercept faculty traffic (e.g., a midterm exam en route to the printer), and that research experiments on the graduate-student VLAN do not inadvertently overload the faculty VLAN.
SIMPLIFYING ACCESS CONTROL POLICIES VLANs provide an effective way to enforce access control by directing inter-VLAN traffic through routers. In addition, by allowing administrators to assign related hosts to IP addresses in the same subnet, VLANs simplify access control configuration by making packet classification rules more concise. Imposing Access Control Policies — VLANs provide a way to restrict communication between hosts. In Fig. 1, router 3 (R3) can apply access control lists (ACLs) to limit the traffic between hosts H3 and H4 that belong to different VLANs. Along the same lines, administrators do not place hosts in the same VLAN unless they
100
Communications IEEE
A
BEMaGS F
are allowed to communicate freely. Campus 3, for example, places all infrastructure services — such as e-mail and DHCP servers — on a single VLAN since these managed services all trust each other. As another example, campus 1 has several “private” VLANs that have no IP router connecting them to the rest of the IP network; for example, the automatic teller machines (ATMs) belong to a private VLAN to protect them from attacks by other hosts. Concise Access Control Lists — Routers and firewalls apply ACLs based on the five-tuple of the source and destination IP addresses, the source and destination TCP/UDP port numbers, and the protocol. Wildcards enable shorter lists of rules for permitting and denying traffic, which simplifies ACL configuration and also makes efficient use of the limited high-speed memory (e.g., TCAMs) for applying the rules. VLANs enable more compact ACLs by allowing administrators to group hosts with common access control policies into a common IP subnet. For example, campus 3 identifies user machines through a small number of IP prefixes (corresponding to the faculty and student VLANs), allowing concise ACLs for traffic sent by user machines (e.g., to ensure only SMTP traffic is allowed to reach the email servers on the infrastructure VLAN). Preventing Source IP Address Spoofing — Source IP address spoofing is a serious security problem, since spoofing allows attackers to evade detection or shift blame for their attacks to others. Assigning host addresses from a common IP prefix simplifies the preventive filtering of packets with spoofed source IP addresses. Hosts in the same VLAN are assigned IP addresses from the same subnet(s). This allows network administrators to configure ACLs at the VLAN’s gateway router to drop any packets with source IP addresses from other prefixes. Campus 3 does precisely that. Supporting Quality of Service — Classifying packets based on IP prefixes applies not only to access control, but also to quality of service (QoS) policies. For example, administrators can configure a router to place IP packets in different queues (with different priority levels) based on the source or destination IP prefix, if hosts are grouped into VLANs based on their QoS requirements. None of the campuses in our study apply these kinds of QoS policies.
DECENTRALIZING NETWORK MANAGEMENT VLANs allow administrators to delegate some management tasks to individual departments. VLANs also simplify network troubleshooting by allowing an administrator to observe connectivity from any part of the campus simply by trunking a port to a VLAN. Federated Management — Campus network administrators sometimes assign all hosts in one department to a VLAN, so each department can have its own control over its hosts in different locations on campus while sharing the same physical infrastructure. Some campuses allocate
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
portions of the VLAN ID space to departments and allow those departments to manage their networks independently. For example, campus 1 has a university-wide IT group and many smaller IT groups. The university-wide group allocates a contiguous block of IP addresses to one VLAN and hands it over to a smaller IT group. One IT group manages a “classroom” VLAN that consists of a computer in each classroom across 60 buildings. Campus 2 allocates a portion of the VLAN ID space to the computer science department and provides a web interface to help the administrators manage the router and firewall settings between the department and the rest of the campus. Campus 4 assigns different gymnasiums across the campus to the same VLAN; administrators for that VLAN can then set firewall rules independently from the rest of the campus. Easier Troubleshooting — VLANs allow network administrators to group hosts based on policy requirements, independent of their locations. If two hosts in the same policy group are in different locations on the campus, administrators can still assign them to the same VLAN so that they can communicate with each other, without interference from intermediate firewalls or routers. In campus 4, the dormitory VLAN spans the campus, including places outside the dormitories; such a setup allows network administrators to help student users diagnose problems since they can put a host on this VLAN anywhere on the campus. Campus 2 also has some VLANs across campus, such as a network-wide VLAN for the IT support team and a VLAN for deploying new experimental management architectures based on OpenFlow [4].
ENABLING HOST MOBILITY VLANs make host mobility easier on a campus wireless network, because hosts can retain their original IP addresses when they move from one access point to another. Allocating a single VLAN to the campus wireless network, as is done in campus 2, allows devices to move anywhere on the campus without having to obtain a new IP address. The campus 2 wireless network has about 6000 active hosts on the same VLAN. These hosts include laptops, mobile phones, passenger counters, and vehicle locators. As users move across the campus on foot or in vehicles, they can remain connected to the campus network, migrating between access points without experiencing disruptions to ongoing connections.
PROBLEM: LIMITED GRANULARITY OF POLICY VLANs are a relatively inflexible way to support policies. In this section, we discuss three main limitations VLANs impose on the granularity of policies — limits on the number of VLANs, limits on the number of hosts per VLAN, and the difficulty of assigning an access port to multiple VLANs without end-host support. We also discuss the incomplete ways administrators try to work around these limitations.
IEEE
BEMaGS F
LIMITED NUMBER OF VLANS VLANs allow
The total number of VLANs is limited because of built-in protocol limitations (i.e., VLAN ID space) and implementation limitations (i.e., switch and router resources): • VLAN ID space: The VLAN ID is a 12-bit header field, limiting a network to 4096 VLANs.1 • Switch memory: Limited memory for storing bridge tables often restricts individual switches to supporting 300–500 VLANs. • Router resources: Inter-VLAN traffic imposes additional load on the routers. Administrators work around these limitations in two ways.
administrators to delegate some management tasks to individual departments. VLANs also simplify network troubleshooting by allowing an administrator to observe connectivity from any part of the
Placing Multiple Groups in the Same VLAN — Administrators can assign multiple groups of hosts to a single VLAN and configure finergrained access control policies at the routers to differentiate between hosts in different groups. Campus 1 combines some groups of hosts together, assigning each group a different block of IP addresses within a larger shared subnet. From the configuration data, we see that about 11 percent of the VLANs have ACLs expressed on smaller IP address blocks. For example, one VLAN contains the DNS servers, logging and management servers, and some dorm network web servers. Although these hosts reside in different locations, are used for different purposes, and have different reachability policies, they are placed in a single VLAN because they are managed by an IT group that has a single VLAN ID and one IP subnet.
campus simply by trunking a port to a VLAN.
Reusing the Limited VLAN Identifiers — To deal with limitations on the number of VLAN IDs, administrators can use the same VLAN ID for multiple VLANs, as long as the VLANs do not have any links or switches in common. Unfortunately, reusing VLAN IDs makes configuration more difficult, since administrators must take care that these VLANs remain disjoint as new hosts, links, and switches are added to the network. Campus 1, in particular, reuses VLAN IDs quite extensively.
LIMITED NUMBER OF HOSTS PER VLAN The overheads of broadcast traffic, flooding, and spanning tree impose limits on the number of hosts in each VLAN. For example, campus 1 has a wireless VLAN with 3000 access points and thousands of mobile hosts that receive a large amount of broadcast traffic. These scalability limitations make it difficult to represent large groups with a single VLAN. Administrators work around this problem by artificially partitioning these larger groups. Dividing a Large Group into Multiple VLANs — A large group can be divided into multiple VLANs. For example, campus 1 has public computer laboratories with 2500 hosts across 16 VLANs. The 1200 hosts in one academic college in campus 1 are divided into eight VLANs. Dividing a large group into multiple VLANs unfortunately prevents mobile hosts from retaining their IP addresses as they move
1
IEEE 802.1QinQ provides a way to extend the ID space using multiple tags.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
101
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
To deal with limitations on the
from one location to another. Additionally, the VLANs must be configured with the same access control policy to retain the semantics that would exist if hosts belonged to a single larger group.
number of VLAN IDs, administrators can use the same VLAN ID for multiple VLANs, as long as the VLANs do not have any links or switches in common. Unfortunately, reusing VLAN IDs makes configuration more difficult.
COARSE-GRAINED ASSIGNMENT OF TRAFFIC TO VLANS Although they are natural for grouping traffic by end host, VLANs are a clumsy way to group traffic across other dimensions (e.g., by application). With end-host support for VLAN tagging, hosts can assign different virtual interfaces to different VLANs. For example, a computer hosting multiple virtual machines can run a software switch that has a different access port (and, hence, can assign a different VLAN) for each virtual interface. However, the end host must support VLANs making it hard to work with the heterogeneous user devices common on college campuses. In addition, the campus administrator must trust the user machine to faithfully apply the appropriate VLAN tag — introducing potential security risks. Although protocols like 802.1x can help authenticate hosts, many campuses do not force all hosts to use these mechanisms. Unexpected problems can arise when administrators assign VLANs directly to access ports. For example, campus 3 assigns each access port to a (single) VLAN dynamically, based on the source MAC address of the attached host. If multiple hosts connect to a single wall jack (e.g., via a common hub or an unmanaged switch), the hosts are assigned to the same VLAN — based on the MAC address of whatever host sends the first packet. Since campus 3 has different VLANs for faculty and students, this can raise security problems when a student plugs into a hub in a faculty member’s office or vice versa. The same problem arises if a single computer runs multiple virtual machines, each with its own virtual interface and MAC address. By connecting to the same switch access port, all of these virtual interfaces would be assigned to the same VLAN, a problem raised by the administrators in campus 2. Restricting each access port to a single VLAN significantly limits the kinds of policies the network can support. For example, administrators cannot assign a single host interface to multiple groups (e.g., a faculty member in the systems group cannot belong to both the faculty VLAN and the systemsgroup VLAN) or have different applications belong to different groups (e.g., web traffic cannot belong to a different VLAN than Skype traffic).
PROBLEM: COMPLEX CONFIGURATION
2
IPv6 might solve the problem but will not be widely deployed in the foreseeable future.
102
Communications IEEE
Although Ethernet was designed with the goal of “zero configuration,” VLAN configuration is challenging and errorprone [5], for two main reasons. First, each host’s IP address must be consistent with the IP subnet of its VLAN. Second, the switches require configuration to ensure each VLAN has an efficient spanning tree that remains connected under common failure scenarios.
A
BEMaGS F
HOST ADDRESS ASSIGNMENT Administrators associate each VLAN with one or more IP subnets and must ensure that the host interfaces within that VLAN are assigned addresses from that block. The tight coupling between VLANs and IP address assignment leads to two problems. Wasting IP Addresses — All four campuses have a one-to-one mapping between an IP subnet and a VLAN. Since IP prefixes must align with power-of-two boundaries, VLANs can lead to fragmentation of the available address space — especially if 5 some VLANs have fewer hosts than others.2 Campus 1, for instance, originally assigned a /24 prefix to each VLAN but, after running out of address space, was forced to use smaller subnets for some VLANs. Complex Host Address Assignment — To ensure that host IP addresses are consistent with the VLAN subnets, Campus 1 manually configures each host with a static IP address from the appropriate VLAN, except for a few VLANs (e.g., the wireless network) that use DHCP. The other campuses use DHCP to automatically assign IP addresses based on the hosts’ MAC addresses. However, the administrators must ensure that DHCP requests reach the DHCP server, even though broadcast traffic only reaches machines in the same VLAN. Rather than devote a DHCP server to each VLAN, campuses 2, 3, and 4 use relay agents to forward requests to a common DHCP server, requiring additional configuration on the routers [6]. Either way, the DHCP server configuration must be consistent with whatever system is used to assign hosts to VLANs.
SPANNING TREE COMPUTATION Switches must be configured to know which VLANs they should support on each trunk link. Administrators must explicitly configure both ends of every trunk link with the list of VLANs to participate in. For example, in Fig. 1, VLAN1 must be allowed on the link between S1 and S2, while VLAN2 need not be permitted. Wrongly omitting a VLAN from that list disrupts communication between the hosts on that VLAN. Unnecessarily including extra VLANs leads to extra broadcast/flooding traffic and larger bridge tables. Determining which links should participate in a VLAN, and which switch should serve as the root bridge of the spanning tree, is often difficult. Limitations of Automated Trunk Configuration — Manual configuration of trunk links is error-prone [7], and inconsistencies often arise as the network evolves [8]. Automated tools, like Cisco’s VLAN Trunk Protocol (VTP) [2], reduce the need for manual trunk configuration. However, these tools require administrators to divide the network into VTP domains, where switches in the same domain cooperate to identify which VLANs each link should support. Each switch must participate in all VLANs in its domain, leading to extra overhead; in fact, some commercial switches can only participate in a handful of
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
spanning-tree instances, limiting the effective size of VTP domains. As a result, campus 1 is divided into several smaller VTP domains, using manually-configured trunk links to interconnect the domains. Campus 2 does not use VTP because some of its switches come from another vendor that does not support Cisco’s proprietary protocol. Campus 3 does not use VTP because the administrators prefer to know by design which links participate in each VLAN, to simplify network troubleshooting. Enabling Extra Links to Survive Failures — Although Ethernet switches can compute a spanning tree automatically, administrators must often intervene to ensure that each VLAN remains connected after a failure. To prevent partitioning of the VLANs, campus 1 installs parallel links between switches and treats them as one logical link; this ensures that the VLANs remain connected even if a physical link fails. To survive switch failures, campus 1 configures the trunk links between the core switches to participate in all VLANs. In general, identifying which links to include is challenging, since enabling too many links in the VLAN is wasteful, but having too few can lead to partitions during failures. Distributing Load Over the Root Bridges — The switches near the root of a spanning tree must carry a large amount of traffic. Dividing the network into multiple VLANs can help distribute the load over multiple spanning trees with different root bridges. By default, the switch with the smallest identifier becomes the root of the spanning tree, resulting in the same switch serving as the root bridge in multiple VLANs. To distribute traffic load more evenly, administrators often configure the root bridge of each VLAN manually. For example, the administrators of campus 1 select the most powerful switches to serve as root bridges.
CONCLUSION We have surveyed four campus networks to better understand and illustrate how VLANs are used in practice. Our analysis indicates that VLANs are used for many objectives that they were not originally intended for, and are often ill-suited for the tasks Further, the use of VLANs complicates network configuration management. We believe future enterprise networks should look at ways to minimize the use of VLANs and explore more direct ways to achieve the network administrators’ objectives with the goal to make management easier for campus and enterprise administrators. To extend our understanding of the VLAN usage in practice, we call for operators of campus and enterprise networks to participate in the survey available at [9].
ACKNOWLEDGMENTS We thank Russ Clark (Georgia Tech), Brad Devine (Purdue), Duane Kyburz (Purdue), Peter Olenick (Princeton), and Chris Tengi (Princeton) for sharing their expertise and experiences about network management and VLANs.
IEEE
BEMaGS F
REFERENCES [1] “VLAN Management Policy Server,” http://www.cisco. com/en/US/tech/tk389/tk689/technologiestech note09186a00800c4548.shtml. ________________ [2] “VLAN Trunking Protocol,” http://www.cisco.com/ e______________________________ n/US/tech/tk389/tk689/technologiestech note09186a0080094c52.shtml. ________________ [3] “Multiple VLAN Registration Protocol,” http://www. cisco.com/en/US/docs/switches/lan/catalyst6500/catos/8. x/configuration/guide/mvrp.pdf. _________________ [4] N. McKeown et al., “OpenFlow: Enabling Innovation in Campus Networks,” ACM Comp. Commun. Rev., Apr. 2008. [5] T. Benson, A. Akella, and D. Maltz, “Unraveling the Complexity of Network Management,” Proc. NSDI, Apr. 2009. [6] C. J. Tengi et al., “autoMAC: A Tool for Automating Network Moves, Adds, and Changes,” Proc. Large Installation Sys. Admin. Conf., 2004. [7] P. Garimella et al., “Characterizing VLAN Usage in an Operational Network,” Proc. Wksp. Internet Network Mgmt., Aug. 2007. [8] X. Sun et al., “A Systematic Approach for Evolving VLAN Design,” IEEE INFOCOM, 2010. [9] http://www.surveymonkey.com/s/X5K5GLM.
We believe future enterprise networks should look at ways to minimize the use of VLANs and explore more direct ways to achieve the network administrators’ objectives with the goal to make management easier for campus and enterprise administrators.
BIOGRAPHIES MINLAN YU (
[email protected]) _______________ is a Ph.D. student in the computer science department at Princeton University. She received her B.S. in computer science and mathematics from Peking University in 2006 and her M.S. in computer science from Princeton University in 2008. She has interned at Bell Labs, AT&T Labs Research, and Microsoft. Her research interest is in network virtualization, and enterprise and data center networks. XIN SUN is a Ph.D. candidate in the School of Electrical and Computer Engineering at Purdue University, West Lafayette, Indiana, where he works with Prof. Sanjay Rao. His research interests are in the design and configuration of large-scale enterprise networks, and the migration of such networks to new architectures. He received his B.Eng. degree in computer engineering from the University of Science and Technology of China in 2005. N ICK F EAMSTER is an associate professor in the College of Computing at Georgia Tech. He received his Ph.D. in computer science from MIT in 2005, and his S.B. and M.Eng. degrees in electrical engineering and computer science from MIT in 2000 and 2001, respectively. His research focuses on many aspects of computer networking and networked systems, including the design, measurement, and analysis of network routing protocols, network operations and security, and anonymous communication systems. In December 2008 he received the Presidential Early Career Award for Scientists and Engineers (PECASE) for his contributions to cybersecurity, notably spam filtering. His honors include the Technology Review 35 “Top Young Innovators Under 35” award, a Sloan Research Fellowship, the NSF CAREER award, the IBM Faculty Fellowship, and award papers at SIGCOMM 2006 (network-level behavior of spammers), NSDI 2005 (fault detection in router configuration), Usenix Security 2002 (circumventing web censorship using Infranet), and Usenix Security 2001 (web cookie analysis). SANJAY RAO is an assistant pProfessor in the School of Electrical and Computer Engineering, Purdue University. He obtained his Ph.D. in computer science from Carnegie Mellon University. His current research interests are in enterprise management and cloud computing. In the past, he has done pioneering work on live streaming using peer-topeer systems. He is a recipient of an NSF Career award, and has served as a Technical Program Chair of the INM/WREN workshop. J ENNIFER R EXFORD is a professor in the Computer Science Department at Princeton University. From 1996 to 2004 she was a member of the Network Management and Performance Department at AT&T Labs–Research. She is coauthor of the book Web Protocols and Practice (Addison-Wesley, May 2001). She received her B.S.E. degree in electrical engineering from Princeton University in 1991 and her Ph.D. degree in EECS from the University of Michigan in 1996.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
103
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
TOPICS IN NETWORK AND SERVICE MANAGEMENT
Toward Fine-Grained Traffic Classification Byungchul Park and James Won-Ki Hong, POSTECH Young J. Won, Internet Initiative Japan
ABSTRACT A decade of research on traffic classification has provided various methodologies to investigate the traffic composition in data communication networks. Many variants or combinations of such methodologies have been introduced continuously to improve the classification accuracy and efficiency. However, the level of classification details is often bounded to identifying protocols or applications in use. In this article, we propose a fine-grained traffic classification scheme based on the analysis of existing classification methodologies. This scheme allows to classify traffic according to the functionalities in an application. In particular, we present a traffic classifier which utilizes a document retrieval technique and applies multiple signatures to detect the peer-to-peer application traffic according to different functionalities in it. We show that the proposed scheme can provide more in-depth classification results for analyzing user contexts.
INTRODUCTION Understanding traffic behavior is an important part of network operations and management. A decade of research on traffic classification has provided various techniques to identify types of traffic information. As the Internet continuously evolves in scope and complexity, its traffic characteristics are also changing in terms of traffic composition and volume. Peerto-peer (P2P) and multimedia traffic applications have rapidly grown in popularity, and their traffic occupies a great portion of the total Internet traffic volume these days. Kim et al. [1] have shown that P2P applications generate a substantial volume in enterprise networks. In 2008, a study by a Japanese Internet service provider (ISP) [2] observed that a significant portion of P2P traffic is recently being replaced by multimedia and web traffic. In particular, a newer generation of P2P applications is incorporated with various obfuscation strategies, such as ephemeral port allocation and proprietary protocols, to avoid detection and filtering. A popular communication application like Skype eludes detection by payload encryption or plain-text ciphers [3]. The dynamic
104
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
nature of Internet traffic adversely affects the accuracy of traffic classification and makes it a more challenging task. The previous studies have discussed various classification methodologies (e.g., well-known port number matching, payload contents analysis, machine learning, etc.). Many variants of such methodologies have been introduced continuously to improve the classification accuracy and efficiency. However, it is extremely difficult for any method to claim 100 percent accuracy due to fast-changing and dynamic nature of the Internet traffic. The classification accuracy is also questionable since there is often no ground truth dataset available. In another respect, each research aims at different levels of classification. Some only had a coarse classification goal such as classifying traffic protocol or application type; while others had more detailed classification goal such as identifying the exact application name. Therefore, it is often unfair to cross-compare each classification method in terms of accuracy. To overcome this issue, we need to investigate how we can provide more meaningful information with such limited traffic classification results rather than focusing on improving 1 or 2 percent of classification accuracy. This article proposes the concept of finegrained traffic classification. A single application typically has several functions and each function triggers a unique traffic characteristics. The finegrained traffic classification can classify various types of traffic, which are generated by a single application. We investigated existing traffic classification studies in terms of classification schemes rather than classification methods. While previous studies focused on classification methods, we have focused on classification output itself. By analyzing the output categories of the other classification research, we propose a new traffic classification scheme. We also present an example of fine-grained traffic classification by applying it to real P2P application traffic. The organization of the article is as follows. We present our related work and our motivation for fine-grained traffic classification. We explain our proposed method, which utilizes a text retrieval technique. We then describe our experiments with the real-world traffic dataset. Finally, concluding remarks and possible future work are discussed.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
BACKGROUND In this section, we describe different traffic classification research according to the level of classification requirements and its analysis capability.
RELATED WORK Application Protocol Breakdown Scheme — Traffic classification is a process of identifying network traffic based on the features that can be passively observed in the traffic. The features and classification results may vary according to specific classification requirements and analysis needs. In early days, traffic classification was performed as part of traffic characterization work, often motivated by the dominance of a certain protocol in a network. Several studies [4, 5] analyzed the packet and byte distributions regarding transport and application layer protocols. TCP/UDP port numbers mapped to a wellknown TCP/UDP protocol. The application protocol breakdown scheme shows a rough estimation of the traffic composition and is still a popular solution at the Internet backbone because of its high and even increasing traffic volumes and limited computing resources for traffic analysis. Borgnat et al. [2] showed that a significant portion of P2P traffic is replaced by multimedia and Web traffic by analyzing longitudinal traffic characteristics of trans-Pacific backbone links. Although they aggregated the P2P traffic with other unknown protocols, they also utilized wellknown port numbers for application protocol breakdown. Traffic Clustering Scheme — Straight-forward classification approaches (e.g., protocol or portbased) cannot provide in-depth classification of similar traffic type generated by different protocols. Traffic clustering scheme refers to traffic workload characteristics rather than protocol traffic decomposition. McGregor et al. [6] proposed a machine learning-based classification method which can break down into clusters: Bulk transfer, small transactions, and multiple transactions. It allows us to understand the major types of traffic in network. Application Breakdown Scheme — The dominance of P2P traffic in the Internet has had a huge influence on traffic classification research and led to more sophisticated heuristics. In this context, many researchers have focused on identifying the exact application represented by the traffic. Discovering byte signatures [7] has been a popular solution. Regardless of its proven accuracy, the signature-based solution possesses high processing overhead and privacy-breaching issues because it requires a packet header and payload inspection. Recently, machine learning techniques which use statistical information of the transport layer [8] are introduced to overcome privacy legislation related to packet payload inspection. They focus on the fact that different applications have different communication patterns (behaviors). Moreover, Szabo et al. [9] introduced combinations of these existing methods in order to balance between the level
of classification completeness and accuracy. All these efforts focused on classifying network traffic according to the name of application in use.
IEEE
BEMaGS F
While the application protocol breakdown scheme cannot
Application-type Breakdown Scheme — BLINC [10] is a connection pattern-based classification method. The idea behind BLINC is to investigate the communication pattern generated by a host and extract behavioral patterns which may represent distinct activities or applications. It categorizes network traffic according to application-type rather than a specific application name, such as Web, game, chat, P2P, streaming, mail, and attack activities. This scheme resides between former two schemes.
distinguish the traffic
MOTIVATION
name represented
Figure 1 shows different traffic classification schemes according to their classification level. The application protocol breakdown scheme presents network traffic into different protocols rather than application types or names. For example, all ftp traffic is classified under the ftp protocol group although there are many distinct ftp client programs since all clients employ the same ftp protocol for data transfer. The traffic clustering scheme was proposed in different perspective to traffic classification. While the application protocol breakdown focuses on identifying certain protocol, the clustering scheme can capture common characteristics shared among the distinct applications using a single or multiple protocols. In addition, the application breakdown scheme can provide more detailed classification results, especially for P2P applications. It would classify distinct application names even if the corresponding traffic is generated from the same protocol. For example, there are many descendant applications which use the BitTorrent protocol. While the application protocol breakdown scheme cannot distinguish the traffic generated by different BitTorrent clients, the application breakdown scheme can classify the traffic according to the exact client name represented by the traffic. The application-type breakdown scheme resides between the traffic clustering and application protocol breakdown schemes in terms of classification level. It characterizes the traffic based on connection pattern or host profiles and classifies into various application-types, such as Web, game, chat, P2P, streaming, mail, and security attack activities. One application-type can be a superset of both application and application protocol. The fine-grained traffic classification can classify various types of traffic which are generated by a single application. As shown in Fig. 1 a single application typically has several functions and each function triggers a unique traffic characteristic. While top n protocol or application analysis is possible with the other schemes, our scheme enables new analysis categories, such as average browsing time to initialize a file download and popular functions in use among users. It is also a tool to analyze user behavior and design future applications in the Internet. When it applies to Web traffic, analyzing the most pop-
by the traffic.
generated by different BitTorrent clients, the application breakdown scheme can classify the traffic according to the exact client
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
105
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
A key to the fined-grained traffic
Function. 2 (e.g., searching)
Function. 1 (e.g., login)
classification is how
Function. k (e.g., downloading)
Fine-grained classification
to categorize single App. j
application traffic into different traffic groups. It is quite
App. protocol n
similar to the traditional traffic classification problem except for the
App. 1
App. m
App. 1
App. i
App. 1
App. j
Application breakdown
degree of classification details.
Application protocol 1
Application protocol 2
Application type 1
Application protocol n
Application protocol breakdown
Application type t
Application-type breakdown
Transport layer
Traffic clustering
IP layer
Figure 1. Traffic classification schemes according to different classification levels. ular function of the Web site (e.g., Facebook) is also possible. This will extend the traffic classification research from network administrative oriented research to user-context-dependent research.
FINE-GRAINED TRAFFIC CLASSIFICATION A key to fine-grained traffic classification is how to categorize single-application traffic into different traffic groups. It is quite similar to the traditional traffic classification problem except for the degree of classification details. Accordingly, various existing methods can be applied to this classification scheme. Most of them use a classifier per application or protocol. We simplify the fine-grained traffic classification problem as follows: to build arbitrary classifiers (e.g., application signature, connection behavior model, statistical model) per application where each classifier corresponds to a distinct function in the application. Among many classifiers, we have selected signature as the classifier because a lot of other previous work has demonstrated that the signaturebased approach is by far the most reliable for accuracy. Even some statistical approaches have used signature as the ground truth for validation [8]. In addition, it is convenient to apply new fine-grained signatures into existing traffic classification systems and commercial traffic shapers, and intrusion detection devices utilize application signatures for their classification .
106
Communications IEEE
Our methodology for fine-grained traffic classification consists of three parts: 1 Input data collection 2 Extraction of fine-grained signatures 3 Traffic classification using fine-grained signatures This article focuses on generating fine-grained classifiers (steps 1 and 2). For signature extraction, we used our previous work, the LASER algorithm [3], which can generate an application signature automatically. Step 1 is not part of the actual signature extraction; however, it is crucial for the entire signature generation process because the input data for the LASER algorithm directly affects the reliability of the signature.
INPUT DATA COLLECTION The LASER algorithm requires sanitized packet collection as its input data. The sanitized raw packets refer to the packets belonging to the target application only. We have developed a continuous packet dump agent using Libpcap to collect the packet trace for every running process in the OS. The collecting agent divides the sanitized packets according to each flow and stores them in a separate packet dump file tagged with the origin process name. It is important to keep the datasets separate according to each flow and process name because it can reduce unnecessary packet comparison overhead in the pattern extraction step. If the given dataset is a mixture of many different applications, it may be difficult to discover a common pattern. Such a design decision is necessary to guarantee the efficiency and accuracy of signature extrac-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
BEMaGS
App. 2
F
The idea behind
Host App. 1
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Application n App. n
Function Function module 1 module 2
document retrieval is Function module m
that the similarity between documents
Login
Network interface
Searching
can be measured by
Downloading
the frequency of keywords in the
Network interface
documents. We have
Traffic dump agent : Traffic
: Process information
: Separated packet dump file Sanitized traffic trace App. 1
App. 2
defined several terms to apply document similarity to traffic classification.
App. n Fine-grained traffic trace Traffic type 1 (e.g., login) Traffic type 2 (e.g., searching)
Fined-grained traffic classifier
Traffic type mi (e.g., downloading)
Figure 2. Input data collection process.
tion; thus, we remove any uncertainty of traffic being fed to the signature extraction algorithm. It increases the possibility of finding a reliable application signature to extract it from similar traffic types in the pool of sanitized traffic. As many functions are embedded in network applications, an application generates different types of traffic according to its function or purpose. For example, a P2P application has various functions such as login, searching, downloading, advertisement, and chatting. In some cases, even Web browsing is included. In order to resolve this issue, we developed a fine-grained flow classifier that could group sanitized flows into several subtypes according to traffic type. Figure 2 illustrates the workflow of the input data collection process. When n different applications are running on a host, each application executes m i different functional modules (the subscript i indicates that the value of mi differs from application to application) and each module in an application generates different types of traffic. The traffic dump agent monitors the network interface continuously and captures all traffic data passing through the network interface. The agent aggregates the traffic data into flows and stores each flow in a separate file. Every flow is tagged with the application or process name acquired from the OS. The stored n groups of sanitized traffic, labeled with an application name, is fed into the fine-grained traffic classifier. The fine-grained traffic classifier classifies the sanitized traffic into mi subcategories according to flow type. Finally, we can get n × mi groups of flow data. Each group of flow data is used as input data for LASER. In this case, a number of LASER’s output is n × mi and a sin-
gle application can have at most mi signatures. Most of the prior research on signature-based traffic classification used a single signature per application. However, this may lead to an increase in the false negative ratio as well. More details on the fine-grained traffic classifier are in the upcoming section.
TRAFFIC CLASSIFIER In order to build the fine-grained traffic classifier, we adopted a document retrieval technique [11] which is one of the main research areas in the natural language processing field. The idea behind document retrieval is that the similarity between documents can be measured by the frequency of keywords in the documents. We have defined several terms to apply document similarity to traffic classification. The following provides our payload vector conversion, vector comparison, and flow comparison methodologies. Payload Vector Conversion — To represent network traffic as a text document, we used vector space modeling (VSM). VSM is an algebraic model which represents text documents as vectors. The objective of document retrieval is to find a subset of documents from a set of stored text documents D that satisfy certain information requests or queries Q. Considering a document space consists of documents Di, each identified by one or more index terms Tj; the terms may be weighted according to their importance [11]. A typical way to determine the significance of a term is measuring the occurrence of the term Tj. When t different index terms are presented in document Di, each document Di is represented
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
107
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
One strength of using Jaccard similarity instead of Euclidean distance is that the similarity value can be normalized and then the similarity calculated by the dot product, and it approaches one if the vectors are similar and zero otherwise.
A
BEMaGS F
1: procedure FLOW GROUPING() 2: FG = {[ ], [ ], u u u , [ ]} // empty flow group (each group consists of flows) // sanitized flows 3: Flow = {f1, f2, u u u , fn} 4: while 1 ) i ) n do 5: if i = 1 then 6: FG[0] @ fi 7: else 8: F = PMF(fi) // convert flow into PFM 9: while 1 ) j ) number of flow group do 10: Similarity{} @ Similarity_Score(FG[j], F) 11: end while 12: if Max(Similarity) * threshold then 13: FG[Max index] @ fi 14: else //Create new flow group 15: FG @ fi 16: end if 17: end if 18: end while 19: return FG 20: end procedure
Algorithm 1. Flow grouping using similarity.
by a t-dimensional term-frequency vector Di = (di1, di2, u u u, dij) where dij represents the frequency of the jth term. While text documents are composed of terms (words), which are units of language as a principal carrier of meaning, a packet does not have basic units containing certain meanings. We have defined the term of a payload as follows to come up with this problem: A term is a payload data within an i-bytes sliding window where the position of the sliding window can be 1, 2, u u u, n – i+1 with n bytes payload. The size of the term set is 28×i, and the length of a term is i. If the word length i is too short, the word cannot reflect the sequence of the byte patterns in the payload. In this case, we cannot recognize the differences among permutations of byte patterns, such as “0 × 01 0 × 02 0 × 03” and “0 × 03 0 × 01 0 × 02.” If the word length is too long, the number of whole representative words increases exponentially. With the definition of term, a packet can be represented as a term-frequency vector called a payload vector. When w i is the occurrence of the i-th term that appears repeatedly in a payload, the payload vector is Payload Vector = [w1w2 u u u wn]T,
(1)
where n is the size of a whole representative term set. We set the sliding window size i to 2 because it is the simplest case for representing the order of content in payloads. When the term size is 2 bytes, the size of all terms is 216. Therefore, the payload vector is represented as a 2 16 -dimensional term-frequency vector. Payload Vector Comparison — Once packets are converted into vectors, the similarity between packets can be calculated by measuring the distance between vectors. We used Jaccard similarity [12] as a distance metric. In our previous work [13], we compared three different similarity metrics: Jaccard similarity, Cosine similarity, and
108
Communications IEEE
RBF. Jaccard similarity showed the best performance without using any sophisticated techniques. The Jaccard similarity J(X, Y) uses word sets from the comparison instances to evaluate similarity. J(X, Y) is defined as the size of the intersection of the word sets divided by the size of the union of the sample sets X and Y: J ( X,Y ) =
X EY X FY
.
One strength of using Jaccard similarity instead of Euclidean distance is that the similarity value can be normalized and then the similarity calculated by the dot product, and it approaches one if the vectors are similar and zero otherwise. If two payload vectors are generated by different applications, the contents of each payload consist of distinct binary sequences and their vectors are also very different. Flow Similarity Comparison — Formula 3 defines the payload flow matrix (PFM). The i-th row of a PFM is the payload vector of the i-th packet in the flow. PFM is a k × n matrix, where k is the number of packets and n is the dimension of payload vectors. Payload flow matrix (PFM) is T PMF = p1 p2 pk , where A pi is the payload vector defined in formula 1. The similarity score between PFMs can be calculated by simple summation addition of the packet similarity values (Similarity Score = Yki=1 J(p i , pvi ), where p i and pvi are i-th packet of the first and second flow accordingly). Algorithm 1 describes the flow grouping process to generate fine-grained flows. The flow grouping procedure reads sanitized flows and groups them into flow groups based on similarity scores. If a flow group set is empty, the first flow f 1 creates a new flow group FG[0] (lines 5–6).
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
5000
7000
3750
5250
Mbytes
Mbytes
BEMaGS F
Signature (application breakdown) Fined-grained traffic classification
Signature (application breakdown) Fined-grained traffic classification
2500
1250
0
A
3500
1750
1
2 Hours
3
0
1
(a)
2 Hours
3
(b)
Figure 3. Traffic volume: fine-grained traffic classification vs. application breakdown: a) Fileguri; b) BitTorrent. Otherwise, the input flow is compared with existing flow groups and inserted into the flow group which has the maximum flow similarity score (lines 10–12). When the maximum similarity score is less than threshold, a new group is created, and flow f i becomes a member of this new group (line 14). Our flow grouping is motivated by an unsupervised machine learning approach since it relies on unlabeled payload vectors to find natural groups, functional clusters in this context. On the contrary, the supervised approach requires pre-labeled datasets to construct a classifier for each cluster. It is difficult to determine the number of functionalities of application in advance. Thus, unsupervised clustering is suitable for fine-grained classifier which intends to identify functional characteristics.
EXPERIMENTS In this section, we provide fine-grained traffic classification results of two representative applications as validation of our proposed classification scheme. We selected P2P applications for verification because of its behavioral (or functional) complexity and popularity in network. We feel that our selection of P2P applications strongly represents the complexity of Internet applications. First, we choose a regionally popular P2P application, called Fileguri, which provides multiple functions — web browsing, searching, downloading, messenger, and commercial advertisement. Second, BitTorrent, a globally favored application especially in Europe and the United States, provides mostly a downloading function. We generated each signature using fine-grained classifier and LASER. To show the advantage of the fine-grained approach, we also analyze the average search counts of user, which cannot be obtained by protocol or application breakdown schemes. For dataset, we collected a full packet trace from our campus network — 3 hours (450 Gbytes) on 16 August, 2007. No port blocking or filtering policy was in effect at the time of measurement.
CLASSIFIER GENERATION PROCESS To generate the input dataset (training data) for fine-grained classifier, we ran our target application while the packet dump agent, described earlier, continuously captures the sanitized trace. Using this sanitized trace as input data, our fine-grained classifier groups flows into clusters. Since there is no ground truth from the perspective of application’s functionality, we manually analyzed flows in each group. For Fileguri, it was possible to determine the functionality by examining the URI fields and requested objects in HTTP “GET” message. After labeling each cluster with functionality, LASER was applied to capture the common patterns shared by clusters. For BitTorrent, the fine-grained classifier grouped the sanitized traffic into nine clusters, which were seemingly many since BitTorrent is simply known for download functionality. We examine the packet payload accordingly to BitTorrent protocol specification and labeled each cluster as downloading, tracker access 1, tracker access 2, distributed hash table (DHT) management 1, DHT management 2, and so on. Note that, DHT managing traffic is not generated by all available BitTorrent clients. It only applies to clients, such as BitTorrent, +Torrent, Transmission, rTorrent, KTorrent, BitComet, and Deluge. However, LASER was not able to generate signatures for one of six DHT clusters. So, we made a simple heuristic to detect the DHT management cluster based on packet size and IP address which generates other BitTorrent traffic as an alternative classifier.
CLASSIFICATION RESULTS We classified our target application traffic using both fine-grained traffic classification and traditional application breakdown methods which use an application signature as a classifier. Figure 3 shows the traffic volume identified by finegrained classification and application breakdown. There is about 10–40 percent difference
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
109
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Downloading Searching Messenger
Web browsing Advertisement
Downloading Tracker access
75
75
Percent
100
Percent
100
20
BEMaGS F
DHT management
20
25
25
0
A
0 1
2 Hours
3
(a)
1
2 Hours
3
(b)
Figure 4. Traffic composition of each functionality: a) Fileguri; b) BitTorrent. where the fine-grained approach discovers more traffic than signature-based application breakdown in each hour. Figure 4 shows the functional decomposition of traffic. Note that we aggregated two tracker access clusters and six DHT management clusters into one functionality for BitTorrent. The downloading portion is dominant (74 and 90 percent) and its volume is close to the volume identified by application breakdown. It implies that application breakdown, which employs signatures, is incapable of detecting other than downloading traffic. Moreover, the web browsing traffic of Fileguri occupies about 12–14 percent, and the same traffic is wrongfully classified as normal HTTP traffic by well-known port matching. It does not even appea under the signature-based method. While previous work has focused on detecting download traffic, it is worthwhile to highlight that the traffic volume of the other traffic in P2P applications is not negligible. The ground truth was verified by the traffic measurement agent (TMA) [3]. It collects process and traffic information in allocation from the host operating system (OS) directly; thus, the information may be the closest possible ground truth available. While verifying the accuracy of fine-grained approach against TMA, there exists a small false positive/negative. We made a few interesting observations on misclassified traffic portions. First, every false positive of Fileguri traffic was caused by unclear boundary to the Web traffic. Although Fileguri provides a limited web browsing function by fixing the “user agent” as Mozilla, users can access the same websites via other Mozilla-based web browsers, such as Firefox, which also sets “user agent” as Mozilla. Second, a false negative is not caused by search or download functionality but an update patch. Both P2P clients update copyrights and prohibited search keywords regularly. We could not easily capture this update traffic for sanitized traffic generation because of its temporal and sporadic communication behavior.
110
Communications IEEE
EXAMPLE OF USER BEHAVIOR ANALYSIS We provide a simple user behavior analysis using fine-grained traffic classification results. With the fine-grained classification results, we can analyze the average search counts when a user initializes downloading in our packet trace. The ratio of searching to downloading in terms of transaction number was 56,392:1. We have empirically confirmed that the Fileguri client generated about 6,000 TCP transactions in a single keyword search. Thus, we conclude that a Fileguri user performs about 9.398 searches on average before downloading from the P2P network. The goal of this simple analysis is to provide the average searching counts of users. However, we believe that the finegrained traffic classification has a much wider application.
CONCLUDING REMARKS Various traffic classification methods have been suggested in order to offer better classification accuracy and information about traffic composition in target networks. In this article, we have proposed a new traffic classification scheme which can classify different traffic types within a single application. In particular, we have presented a fine-grained traffic classifier which utilizes a text retrieval technique and applies multiple signatures to detect P2P traffic according to different functionalities. Our proposed scheme can provide more in-depth classification results for analyzing user contexts. It also benefits network operators who need to view the detailed traffic composition of network, and researchers who want to study the user behavior. For future work, we plan to analyze the flexibility of our approach by applying different classification methodologies instead of multiple signatures. We also plan to conduct various user behavior and context analysis based on finegrained traffic classification.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
ACKNOWLEDGMENTS This research was supported by the World Class University (WCU) program through the National Research Foundation of Korea funded by the Ministry of Education, Science and Technology (R31-2010-000-10100-0) and the Korea Communications Commission (KCC) under the Novel Study on Highly Manageable Network and Service Architecture for the New Generation support program supervised by the Korea Communications Agency (KCA; KCA-201110921-05003).
REFERENCES
BIOGRAPHIES BYUNGCHUL PARK [S] (
[email protected]) ___________ received his B.Sc. degree in computer science from POSTECH, Korea, in 2006. He is a Ph.D. student in the Department of Computer Science and Engineering, POSTECH. His research interests include Internet traffic measurement and analysis, and intelligent traffic classification.
F
YOUNG J. WON [M] (
[email protected]) _________ is a researcher at IIJ Research Laboratory, Tokyo, Japan. Prior to IIJ, he was a postdoctoral researcher at INRIA, France. He received his B.Math (2003) from the University of Waterloo, Canada, and M.S. (2006) and Ph.D. (2010) from POSTECH. JAMES WON-KI HONG [SM] (
[email protected]) _____________ is a professor and head of the Division of IT Convergence Engineering at POSTECH. He received a Ph.D. degree from the University of Waterloo in 1991. His research interests include network management, network monitoring and analysis, convergence engineering, ubiquitous computing, and smartphonomics. He has served as Chair (2005–2009) of the IEEE ComSoc Committee on Network Operations and Management (CNOM). He is serving as Director of Online Content for IEEE ComSoc. He is a NOMS/IM Steering Committee Member and a Steering Committee Member of APNOMS. He was General Chair of APNOMS 2006, and General Co-Chair of APNOMS 2008 and APNOMS 2011. He was General Co-Chair of IEEE/IFIPS NOMS 2010. He is an Associate Editor-in-Chief of IJNM and an editorial board member of IEEE TNSM, JNSM, JCN, and JTM.
IEEE Communications Magazine • July 2011
IEEE
BEMaGS
[9] G. Szabó, I. Szabó, and D. Orincsay, “Accurate Traffic Classification,” IEEE WOWMOM 2007, Helsinki, Finland, June 18–21, 2007, pp. 1–8. [10] T. Karagiannis, K. Papagiannaki, and M. Faoutsos, “BLINC: Multilevel Traffic Classification in the Dark,” ACM SIGCOMM 2005, Philadelphia, PA, USA, Aug. 22–26, 2005, pp. 229–40. [11] G. Salton, A. Wong, and C.-S. Yang, “A Vector Space Model for Automatic Indexing,” Commun. ACM, vol. 18, no. 11, 1975, pp. 613–20. [12] L. Hamersa et al., “Similarity Measures in Scientometric Research: The Jaccard Index Versus Salton’s Cosine Formula,” Info. Processing and Mgmt.: An Int’l. Journal, vol. 25, no. 3, May 1989, pp. 315–18. [13] J. Y. Chung et al., “An Effective Similarity Metric for Application Traffic Classification,” IEEE/IFIP NOMS 2010, Osaka, Japan, Apr. 19–23, 2010, pp. 286–92.
[1] M.-Sup Kim, Y. J. Won, and J. W. Hong, “Characteristic Analysis of Internet Traffic from the Perspective of Flows,” J. Comp. Commun., vol. 29, no. 10, June 19, 2006, pp. 1639–52. [2] P. Borgnat et al., “Seven Years and One Day: Sketching the Evolution of Internet Traffic,” IEEE INFOCOM 2009, Rio de Janeiro, Brazil, Apr. 19–25, 2009, pp. 711–19. [3] B.-C. Park et al., “Towards Automated Application Signature Generation for Traffic Identification,” IEEE/IFIP NOMS 2008, Salvador, Bahia, Brazil, Apr. 7–11, 2008, pp. 160–67. [4] K. Thompson, G. J. Miller, and R. Wilder, “Wide-Area Internet Traffic Patterns and Characteristics,” IEEE Network, vol. 11, no. 6, 1997, pp. 10–23. [5] D. Moore et al., “The CoralReef Software Suite as a Tool for System and Network Administrators,” 15th USENIX Conf. System Administration, San Diego, CA, USA, Dec. 2001, pp. 133–44. [6] A. McGregor et al., “Flow Clustering Using Machine Learning Techniques,” PAM Wksp. 2004, Antibes Juanles-Pins, France, Apr. 19–20, 2004, pp. 205–14. [7] S. Sen, O. Spatscheck, and D. Wang, “Accurate, Scalable In-Network Identification of P2P Traffic Using Application Signatures,” WWW Conf. 2004, New York, NY, USA, May 17–20, 2004, pp. 512–21. [8] H. Kim et al., “Internet Traffic Classification Demystified: Myths, Caveats, and the Best Practices,” ACM CoNEXT Conf., Madrid, Spain, Dec. 9–12, 2008, pp. 1–12.
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
111
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
SERIES EDITORIAL
SOME RECENT IEEE STANDARDS FOR WIRELESS DATA COMMUNICATIONS
Mostafa Hashem Sherif
W
ireless communication would not be possible without the radio spectrum as a transport medium. Like air, water or oil, this natural resource needs careful management, including frequency licensing, release and reallocation. The following articles provide a glance at the work many IEEE standards committees have undertaken to satisfy demands for reliable and high-throughput wireless data communication by taking advantage of recent regulatory changes, mostly in the unlicensed part of the spectrum. The first article by Baykas et al. is entitled “IEEE 802.15.3c: The First IEEE Wireless Standard for Data Rates over 1 Gb/s.” It presents the specifications approved in 2009 for personal area networks transmitting at rates exceeding 1 Gb/s and operating in the unlicensed 60 GHz (millimeter) wave band. In the 1990s, regulatory bodies made this band available for use worldwide. The standard divides the frequency band into four common global channels and defines three physical layers, each with its data rate, modulation scheme, and error correction mechanism. An optional codebook-based beamforming protocol was agreed to extend the range of communication. The protocol is independent of the physical layer and is applicable to many antenna configurations. Ancillary activities included the development of a new 60 GHz indoor channel model that takes into account the effects of smaller wavelength and high directivity. Another is the development of a frame aggregation method with low latency suitable for the high data rates of transmission of very short commands, such as in the case of bidirectional communication between a PC and its peripherals. The next contribution by Ying Li et al., “Overview of Femtocell Support of Advanced WiMAX Systems,” concerns the use of femtocells in the next generation of wireless metropolitan networks. Femtocells give wireless broadband access in the licensed part of the spectrum to a service provider’s network on a plug-and-play basis. The
112
Communications IEEE
Yoichi Maeda
article highlights the main agreements made in 2009 and 2010 concerning the evolution of the WiMAX architecture in the IEEE and the WiMAX Forum as documented in the 2011 version of IEEE 802.16m (Part 16: Air Interface for Broadband Wireless Access Systems — Advanced Air Interface). In particular, the article details the recent development for femtocell design in the area of reliability and interference management. A new generation of utility networks (for electricity, water, natural gas, and sewers) could use telemetry and data communication to improve operational efficiency. Because these networks are associated with the smart grid concept, they are often called smart utility networks (SUNs). In the final article, “Smart Utility Networks in TV White Space,” Sum et al. evaluate possible ways that TV white space could be used for communication in SUNs. TV white space is the spectrum of radio frequencies located between existing TV stations that can be released for broadcast TV or low-power wireless devices to provide unlicensed communication. The authors advocate the use of a global set of operating frequencies for SUN usage so that products could interoperate on a worldwide basis. Among the relevant IEEE standardization projects they discuss is the one by Task Group 802.15.4g to ensure SUN interoperability in regionally available unlicensed frequency bands. The scope of Task Group 802.11ah is the unlicensed bands below 1 GHz for smart grid and smart utility communications. The activities of IEEE 802.11af concern modifications to the physical and medium access control (MAC) layers of 802.11 to enable communications in TV white space. IEEE 802.22 addresses the use of cognitive radio to share resources in TV white space from the perspective of a regional area network. Finally, IEEE Technical Advisory Group (TAG) 802.19 is considering various “coexistence scenarios” to allow network elements and TV devices to share the frequency band without interference. The articles required several revisions to reach the
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
SERIES EDITORIAL appropriate level of detail that would benefit the readers of IEEE Communication Magazine. The editors would like to thank the authors for their cooperation. In addition, the 19 reviewers listed below in alphabetical order gave excellent comments that shaped and improved the initial submissions: Bader, Faouzi, Centre Tecnologic de Telecomunicacions de Catalunya — CTTC, Spain Bongartz, Harald, Fraunhofer Institute for Communication, Information Processing and Ergonomics (FKIE), Germany Chang, Kuo-Hsin, Elster Solutions, United States Comstock, David, Huawei, United States Ding, Gang, Olympus Communication Technology of America, United States Eichen, Elliott, Massachusetts Institute of Technology (MIT), United States Garroppo, Rosario, University of Pisa, Italy He, Xingze, University of Southern California, United States Howie, John, Microsoft, United States Liu, Dake, Linköping University, Sweden Lohier, Stéphane, Université Paris-Est, France Pucker, Lee, Wireless Innovation Forum, United States Shyy, D. J, MITRE, United States Singh, Summit, University of California, Santa Barbara, United States Tarchi, Daniele, University of Florence, Italy
Uslar, Mathias, OFFIS (Institute for Information Systems), Germany Vardhe, Kanchan, Illinois Institute of Technology, United States Web, William, Ofcom, United Kingdom Wolff, Richard, Montana State University, United States
BIOGRAPHIES MOSTAFA HASHEM SHERIF (
[email protected]) __________ has been with AT&T in various capacities since 1983. He has a Ph.D. from the University of California, Los Angeles, an M.S. in the management of technology from Stevens Institute of Technology, New Jersey, and is a certified project manager of the Project Management Institute (PMI). Among the books he has authored are Protocols for Secure Electronic Commerce (2nd ed., CRC Press, 2003), Paiements électroniques sécurisés (Presses polytechniques et universitaires romandes, 2006), and Managing Projects in Telecommunication Services (Wiley, 2006). He is a co-editor of two books on the management of technology published by Elsevier Science and World Scientific Publications in 2006 and 2008, respectively, and is the editor of the Handbook of Enterprise Integration (Auerbarch, 2009). Y OICHI M AEDA [M] (
[email protected]) ________________ received B.E. and M.E. degrees in electronic engineering from Shizuoka University, Japan, in 1978 and 1980, respectively. Since joining NTT in 1980, for the last 30 years he has been engaged in research and development on access network transport systems for broadband communications including SDH, ATM, and IP. From 1988 to 1989 he worked for British Telecom Research Laboratories in the United Kingdom as an exchange research engineer. He currently leads the Japanese telecommunication standardication organization, the Telecommunication Technology Committee (TTC), since October 2010. In October 2008 at the World Telecommunication Standardization Assembly (WTSA-08), he was appointed chair of ITU-T SG15 for the 2009–2012 study period for his second term. He is a Fellow of the IEICE of Japan. He has been a Series Editor of the Standards Series in IEEE Communications Magazine since 1999.
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
113
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
TOPICS IN STANDARDS
IEEE 802.15.3c: The First IEEE Wireless Standard for Data Rates over 1 Gb/s Tuncer Baykas, Chin-Sean Sum, Zhou Lan, Junyi Wang, M. Azizur Rahman, and Hiroshi Harada, NICT Shuzo Kato, NICT and Tohoku University
ABSTRACT This article explains the important features of IEEE 802.15.3c, the first wireless standard from IEEE in the 60-GHz (millimeter wave) band and its development. The standard provides three PHY modes for specific market segments, with mandatory data rates exceeding 1 Gb/s. During the span of the standard development, new contributions to wireless communication technology were also made, including a new channel model, a codebook-based beamforming scheme, and a low-latency aggregation method.
INTRODUCTION Wireless system designers dream of replacing all cables for indoor data communication with high-speed wireless connections. Unfortunately, the dedicated unlicensed frequency spectrum for this purpose was insufficient until the U.S. Federal Communications Commission (FCC) declared that the 57–64 GHz band could be used [1]. Japan, in turn, allocated the 59–66 GHz band. With the latest adoption of the European Telecommunications Standards Institute (ETSI) 57–66 GHz band, there is now a common continuous 5 GHz band available around 60 GHz in most of the major markets (Fig. 1). As the wavelength of a signal at 60 GHz is around 5 mm, it is called the millimeter wave (mmWave) band. Another important breakthrough was the introduction of relatively cheap and power-efficient complementary metal oxide semiconductor (CMOS) processing for semiconductor manufacturing of 60 GHz band devices. As a result, the price and power requirements for consumer devices were met. For successful commercialization, the final need for developers was a standard that would support almost all usage models. The IEEE 802 LAN/MAN Standards Committee has many success stories in developing global wireless standards, such as 802.11 (WiFi) and 802.15.4 (Zigbee). Within IEEE 802, interest in developing an mmWave physical layer (PHY) began in July 2003, with the formation of an interest group under the 802.15 working group for wireless personal area networks
114
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
(WPANs). According to the IEEE 802 procedure, if the interest group is successful, it is followed by a study group, which decides the scope of the new standard. In March 2004, a study group for mmwave PHY was formally created. The study group members agreed that development of a new PHY to transmit 1 Gb/s or higher data rates is feasible. It was decided to reuse an existing medium access control (MAC) layer (IEEE 802.15.3b), with necessary modifications and extensions. After the approval of the project authorization request, a task group was created. The task group first focused on creating usage models, a 60 GHz indoor channel model, and evaluation criteria. After two years of hard work, three PHY modes and multiple MAC improvements were selected to support different usage models. After various letter ballots, sponsor ballots, and resulting improvements, in September 2009 the IEEE-SA Standards Board approved IEEE 802.15.3c-2009 [2]. It took four and a half years for the task group to complete the standard. Such a duration has been common for many IEEE standards that provide new PHYs. The rest of this article explains the salient features of the standard, as well as some important outcomes. The organization of the article follows the order of the task group’s standardization process. First, we provide the details of the usage models and channel model, which were finalized before other topics. Afterward, channelization is explained, which is common for all PHY modes. Details of the three different PHY modes, new MAC layer features, and beamforming procedures are explained in the following sections, respectively. Finally, the last section presents the conclusions.
USAGE MODELS OF 802.15.3C During the beginning of the standardization process, the 802.15.3c Task Group conducted a detailed analysis of the possible consumer applications in the 60 GHz band. A total of five usage models (UMs) were accepted by the group [3]. UM 1) Uncompressed video streaming: The particularly large bandwidth available in the 60 GHz band enables sending HDTV signals, thus
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
eliminating the need for video cables from highdefinition video players to display devices. The 802.15.3c Task Group assumed 1920 × 1080 pixel resolution and 24 b/pixel for video signals. Assuming a rate of 60 frames/s, the required data rate is found to be over 3.5 Gb/s. The operation range should be 10 m with a pixel error rate below 10–9. UM 2) Uncompressed multivideo streaming: In some applications, multiple video signals could be supplied by a single transmitter. For example, a TV and a DVD recorder could receive signal from a single set-top box, or a TV can display two TV channels side by side on the same screen. According to this UM, the 802.15.3c system should be able to provide video signals for at least two 0.62 Gb/s streams. This data rate corresponds to a video signal with 720 × 480 pixels/frame. UM 3) Office desktop: In this UM, it is assumed that a personal computer communicates with external computer peripherals, including printers, display devices, and hard disks, unidirectionally or bidirectionally. The model mainly assumes data communications, where retransmissions are possible. A target packet error rate of 0.08 is assumed. UM 4) Conference ad hoc: This UM considers a scenario where many computers are communicating with each other using one 802.15.3c network. Most communications are bidirectional, asynchronous, and packet-based. The conference ad hoc UM requires longer ranges than the office desktop UM for improved quality of service. UM 5) Kiosk file downloading: In the last UM, the task group assumed electronic kiosks that enable wireless data uploads and downloads with their fixed antennas. Users will operate handheld devices, such as cell phones and cameras with low-complexity, low-power transceivers. One possible application is downloading video and music files. This model requires 1.5 Gb/s at 1 m range. Figure 2 illustrates the uncompressed video streaming, office desktop, and kiosk file downloading UMs.
60 GHZ CHANNEL MODEL When compared with other indoor wireless systems at 2.4 and 5 GHz, 60 GHz systems have much smaller wavelength and thus the potential for higher directivity. The conventional SalehValenzuela (S-V) channel model [4], which fits the non-line-of-sight (NLOS) communications of previous IEEE 802.11 and IEEE 802.15 specifications, does not suit 60 GHz applications well. A new channel model was accepted by the IEEE 802.15.3c channel modeling subcommittee. This new model combines a line-of-sight (LOS) component using a two-path model with the NLOS reflective clusters of the S-V model [5]. Figure 3 illustrates the new model, including some of the important parameters used to define it. The group decided on a set of eight different parameters to cover all the possible scenarios, as shown in Fig. 3. In addition, path loss coefficients and shadowing values have been extracted for different scenarios.
2160 MHz 1760 MHz
240 MHz
IEEE
BEMaGS
1 58
2 59
60
57 GHz
57 GHz
62
USA, KOREA 59 GHz
4
3 61
F
120 MHz
63
64
65
fGHz
64 GHz
JAPAN
66 GHz
EUROPE
66 GHz
Figure 1. Channelization of 802.15.3c and unlicensed bands around the globe.
CHANNELIZATION OF 802.15.3C One of the main challenges during standardization was providing a common channelization for all the available bands around the world. The bands should be large enough to support the data rates required in the usage models with robust low-spectrum-efficiency modulation schemes. Common global channels should be created for early market penetration. Lower and upper guard bands should also be large enough to minimize interference with other bands. These requirements were satisfied by allocating four bands of 2160 MHz, as shown in Fig. 1. Channelization allows two common global channels, channels 2 and 3. Devices in the United States can use channels 1–3, and those in Japan can use channels 2–4. The lower and upper guard bands have been set as 240 and 120 MHz, respectively. Although they are not equal, the guard bands are wide enough to reduce out-ofband emissions. Following IEEE 802.15.3c, three other standardization bodies, the European Computer Manufacturers Association (ECMA), WirelessHD, and IEEE 802.11 TGad, accepted the same channelization. This is extremely important for the coexistence of heterogeneous 60 GHz systems, because it simplifies detection of other systems and avoidance.
PHY LAYER DESIGN IN 802.15.3C Due to conflicting requirements of different UMs, three different PHY modes have been developed: • Single carrier mode of the mmWave PHY (SC PHY) • High-speed interface mode of the mmWave PHY (HSI PHY) • Audio/visual mode of the mmWave PHY (AV PHY) The SC PHY is best suited for kiosk file downloading (UM5) and office desktop (UM3) usage models. The HSI PHY is designed mainly for the bidirectional, NLOS, low-latency communication of the conference ad hoc model (UM4). The AV PHY is designed to provide high throughput for video signals in video streaming usage models (UM1, UM2). A comparison of the different PHY modes is given in Table 1. The main difference between the different PHYs is the modulation scheme. The SC PHY uses single carrier modulation, whereas the AV
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
115
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
In the receiver for the LOS environment, conventional matched filtering is sufficient for achieving acceptable performance. For the NLOS environment, such as the frequency domain equaliza-
(b)
(a)
optional methods
(c)
Figure 2. Usage models for 60 GHz applications: a) uncompressed video; b) office desktop; c) kiosk downloading.
included to mitigate
Relative power
σ1
Γ: Cluster decay factor 1/Λ: Cluster arrival rate γ: Ray decay factor 1/λ: Ray arrival rate σ1: Cluster lognormal standard deviation σ2: Ray lognormal standard deviation σϕ: Angle spread of rays within cluster (Laplace distribution) Ω0: Average power of the first ray of the first cluster
σ2
An gle of a
multipath fading.
rriv al
tion (FDE) may be
Γ γ
Ω0
σϕ
1/Λ
1/λ
Time of arrival
Figure 3. Graphical representation of the 60 GHz channel model [6].
PHY and HSI PHY use the orthogonal frequency-division multiplexing (OFDM) modulation. In SC modulation, one symbol occupies the whole frequency band, and thus its duration is very short. In OFDM, the available frequency band is divided into orthogonal subcarriers, and data symbols are sent using those subcarriers. In general, SC modulation allows lower complexity and low power operation, whereas OFDM suits well in high spectral efficiency and NLOS channel conditions. Orthogonality of the subcarriers in OFDM allows the use of inverse fast Fourier transform (IFFT) at the transmitter and fast Fourier transform (FFT) at the receiver. All PHY modes have a typical signal frame format consisting of a preamble, header, and payload. The preamble is used for frame detection, channel estimation, frequency recovery, and timing acquisition. The header contains essential information such as payload size, modulation, and coding used in the payload. The payload includes the data to be transmitted.
SINGLE CARRIER MODE OF THE MMWAVE PHY The SC PHY provides three classes of modulation and coding schemes (MCSs) focusing on different wireless connectivity applications. Class 1 is specifically designed for addressing kiosk file
116
Communications IEEE
downloading and the low-power, low-cost mobile market with data rates of up to 1.5 Gb/s. Class 2 is specified for the office desktop UM achieving data rates up to 3 Gb/s, and class 3 is specified for supporting high-performance applications with data rates exceeding 3 Gb/s. Regarding the modulation schemes used in the SC PHY, the support of //2-shifted binary phase shift keying (//2 BPSK) is mandatory for all devices. It is achieved by adding anti-clockwise //2 rotation to the BPSK constellation after each symbol transmission. Added rotation reduces spectrum regrowth due to power amplifier nonlinearities and improves peak-to-average power ratio. Another important point is that //2 BPSK can be realized by minimum shift keying (MSK) modulation with proper filtering. Other supported modulation schemes are //2 quadrature PSK (QPSK), //2 8-PSK, and //2 16-quadrature amplitude modulation (QAM), on-off keying (OOK), and dual alternate mark inversion (DAMI). There are two main forward error correction (FEC) schemes specified in the standard, ReedSolomon (RS) block codes and low-density parity check (LDPC) block codes. RS codes are selected for their low complexity in high-speed communication. RS(255,239) is the main FEC in the system and is used for payload protection.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
F
SC PHY
HSI PHY
AV PHY
Main usage model
Kiosk downloading, office desktop
Conference ad hoc, office desktop
Video streaming, multivideo streaming
Data rates
0.3 Mb/s–5.28 Gb/s
1.54–5.78 Gb/s
0.95–3.8 Gb/s
Modulation scheme
Single carrier
Orthogonal frequency-division multiplexing
Orthogonal frequency-division multiplexing
Forward error control coding options
Reed Solomon code, low-density parity check codes
Low-density parity check codes
Reed Solomon code, convolutional coding
Block size/fast Fourier transform size
512
512
512
Table 1. Comparison of the three modes of mmWave PHY.
HIGH-SPEED INTERFACE MODE OF THE MMWAVE PHY As mentioned earlier, the HSI PHY is designed mainly for computer peripherals that require low-latency bidirectional high-speed data, focusing on the conference ad hoc UM, and uses OFDM. The FFT size is selected as 512, which is necessary in the 60 GHz channel. As OFDM modulation has an inherent complexity due to IFFT and FFT operations, only the LDPC coding scheme is used in the HSI PHY, which offers better coding gains than RS coding. Four FEC rates are obtained using LDPC(672,336), LDPC(672,504), LDPC(672,420), and LDPC(672,588) codes. In terms of modulation, three modulation schemes are selected: QPSK, 16-QAM, and 64QAM. The highest PHY SAP data rate is 5.775 Gb/s. In this mode, a special low-data-rate modulation and coding scheme is created by applying a spreading factor of 48, which can only be used for beaconing and control signals.
45
Modulation and coding schemes with LDPC coding Modulation and coding schemes with RS coding
40 35 30 Range (m)
Four LDPC coding schemes (LDPC(672,336), LDPC(672,504), LDPC(672,588) and LDPC (1440.1344)) with different coding rates are specified to provide higher coding gain with reasonable implementation complexity. In Fig. 4, we provide a performance comparison between RS and LDPC codes in terms of data rate and range in an LOS environment. We have assumed a transmit power of 10 dBm, antenna gain of 10 dBi, and MCSs with //2 BPSK and //2 QPSK modulation at a packet error rate (PER) of 0.08. The results indicate that with LDPC codes, it is possible to increase the range up to 1.5 times at lower data rates. Spreading with either linear feedback shift register code or Golay sequence is also applied to further increase system robustness. The possible spreading values are 2, 4, and 64. In the receiver for the LOS environment, conventional matched filtering is sufficient for achieving acceptable performance. For the NLOS environment, optional methods such as the frequency domain equalization (FDE) may be included to mitigate multipath fading.
25 20 15 10 5
0
500
1000
1500 2000 Data rate (Mbps)
2500
3000
Figure 4. Range vs. data rates of several SC PHY mode modulation and coding schemes. Schemes with LDPC coding use //2 BPSK with spreading factors 2 and 1, and //2 QPSK. Schemes with RS coding use //2 BPSK with spreading factors 4, 2, and 1, and //2 QPSK.
AUDIO/VISUAL MODE OF THE MMWAVE PHY As video and audio devices could be designed only as a wireless data source (e.g., DVD player) or only as a data sink (e.g., HDTV), highly asymmetric data transmission is possible; hence, the designers of the AV PHY mode created two different sub-PHY modes: high-rate PHY (HRP) for video transmission and low-rate PHY (LRP) for the control signal. Both of the sub-PHY modes use OFDM. The HRP mode has an FFT size of 512 and uses all the channel bandwidth available. There are three MCSs with equal error protection, delivering data rates of 0.952, 1.904, and 3.807 Gb/s. There are two MCSs with unequal error protection and two MCSs, in which only the most significant bits are sent. On the other hand, the LRP mode occupies only 98 MHz bandwidth, and three LRPs are arranged per HRP channel. This allocation is to accommodate three different networks in one
IEEE Communications Magazine • July 2011
Communications IEEE
3500
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
117
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
MSDU#3
MSDU#2
A
BEMaGS F
MSDU#1
Step 1: Fragmentation MSDU#3 fragment #2
MSDU#3 fragment #1
MSDU#2 fragment #2
MSDU#2 fragment #1
MSDU#1
Step 2: Mapping Sh#1
Sh#2
MSDU#1
Sh#3
Subframe#2
Sh#4
Subframe#3
MSDU#2 Fragment #1
FCS
Subframe#4
MSDU#2 Fragment #2
FCS
FCS
FCS
MSDU#3 Fragment #1
MAC header
Subframe#1
Step 3: PHY to transmit Subframe#2
Subframe#1
MAC subheaders
HCS
Subframe#3
HCS
Subframe#4
MAC header PHY header
(a)
MSDU#3
MSDU#2
MSDU#1
Step 1: Mapping
Subframe#2
MSDU#1
Sh#1
Subframe#3
MSDU#2
FCS Sh#2
Subframe#4
Zero length MSDU
FCS
Sh#3
FCS
MSDU#3
MAC subheader
MAC header
Subframe#1
Step 2: PHY to transmit Subframe#2
Subframe#1
MAC subheader
HCS
Subframe#3
HCS
Subframe#4
MAC header PHY header
(b)
Figure 5. Aggregation methods in 802.15.3c: a) standard; b) low-latency.
channel, because the HRP modes are assumed to have high beamforming gains. The AV PHY uses RS code as the outer code and convolutional coding as the inner code in the HRP mode, whereas only convolutional coding is used in the LRP mode. Modulation schemes used in the AV PHY are limited to QPSK and 16-QAM.
118
Communications IEEE
both header and payload using a Golay sequence of length 64 chips. The CMS preamble is carefully designed to provide good performance for synchronization and channel estimation, even in poor channel conditions.
COMMON-MODE SIGNALING
MAC LAYER ENHANCEMENTS OF IEEE 802.15.3C
Common mode signaling (CMS) is a low-datarate SC PHY MCS, designed to mitigate interference among different PHY modes. CMS is a common platform that enables different PHY modes to communicate with each other before conducting their respective data transmissions. CMS is used for transmission of the beacon frame, synchronization (sync) frame, and other important command frames, such as the association frame and beamforming training sequence frames. The modulation for CMS is //2 BPSK. The payload part of CMS is encoded with RS(255,239). The FEC for the CMS header is RS(33,17), which is a shortened version of RS(255,239). The code spreading is applied to
Before going into the details of the MAC layer enhancements, we briefly introduce the IEEE 802.15.3c MAC. It is based on the IEEE 802.15.3b standard, which itself is an improvement over IEEE 802.15.3. In the standard a network is called a piconet, which is formed in an ad hoc fashion. Among a group of devices (DEVs), one will act as the piconet coordinator (PNC) to provide the piconet’s synchronization and to manage access control of the rest of the DEVs. The necessary control information is embedded in beacons. Upon receiving a beacon from a PNC, the DEVs become aware of the existence of the piconet. Beacons provide information about when and how DEVs can access the network.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
During network operation, time is divided into sequential superframes (SFs). Each SF has three segments: a beacon period, a contention access period (CAP), and a channel time allocation period (CTAP). During the beacon period, the PNC sends one or multiple beacons. The CAP is reserved mainly for command and control communication between PNC and DEVs. Since such a communication is mainly asynchronous, a suitable access method is selected, carrier sense multiple access with collision avoidance (CSMA/CA). The remaining time of an SF includes the CTAP, which provides time-division multiple access (TDMA) communications. The CTAP is composed of multiple channel time allocations (CTAs). Each CTA is a time slot granted by the PNC for a certain pair of DEVs. Time-sensitive applications such as AV streaming use the CTAP for guaranteed data transmission. With these specifications, system designers had already developed an efficient and well structured MAC layer that only required improvements in three major areas: • Providing coexistence among different PHYs and avoiding interference from hidden devices • Improving transmission efficiency to enable MAC SAP rates over 1 Gb/s and providing low-latency transmissions for delay-sensitive applications • Supporting directivity inherent to 60 GHz signals and beamforming antennas In the next two subsections, we explain enhancements in the first two areas. The solution in the last area requires a complex beamforming algorithm, which is explained later.
COEXISTENCE AMONG 802.15.3C PHYS To achieve better coexistence among DEVs using different PHY modes, a sync frame is introduced. A sync frame includes information about the duration of the SF and timing information of the CAP and each CTA. Sync frames are modulated using the CMS mentioned in the previous section. According to 802.15.3c rules, it is mandatory for all PNC-capable DEVs to transmit a sync frame in every SF. In addition, any PNC-capable DEV shall be able to receive and decode sync frames and other command frames modulated with CMS. As a result, any PNC-capable DEV, regardless of its PHY mode in operation, will be informed about the existence of nearby piconets. It will then have the opportunity to join one instead of starting another independent piconet. The sync frame transmission can thus be seen as an effective coexistence method to mitigate potential co-channel interference from other piconets. Apart from the rules for PNC-capable DEVs, an optional rule related to non-PNC-capable DEVs is also defined in the standard: Any DEV capable of transmitting a sync frame may do so in the first granted CTA in an SF and in every predefined number of SFs. This rule is intended to further extend the coverage area of the sync frame. It allows nonPNC-capable DEVs to participate in the sync frame transmission.
IEEE
BEMaGS F
FRAME AGGREGATION In high-speed WPAN and WLAN systems, transmission efficiency decreases with the increase in transmission speed due to the increased ratio of overhead time to payload transmission time. To improve transmission efficiency and throughput performance, frame aggregation can be employed. The basic idea of frame aggregation is to reduce the overhead, such as the preamble and PHY/MAC header, by concatenating multiple MAC service data units (MSDUs) to form a frame with a long payload. In the IEEE 802.15.3c standard, two novel aggregation methods are specified: standard aggregation and low-latency aggregation. Standard Aggregation — Standard aggregation is designed to support transmission of uncompressed video streaming. Figure 5a shows the basic procedure of standard aggregation. The MAC layer of the transmitter, upon receiving an MSDU from the upper layer, divides the MSDU into small pieces of data blocks if the length of the MSDU exceeds a predefined threshold. This process is called fragmentation. The MAC attaches a frame check sequence (FCS) to each data block to form a subframe. For each subframe, there is a subheader (Sh) created to carry information needed for the receiver to decode individual subframes, such as subframe length, MSDU sequence number, and used MCS. The MAC header, on the other hand, carries high-level control information applicable to all the subframes, such as source and destination addresses. All the subheaders are placed back-to-back and attached to a single header check sequence (HCS) to form a MAC subheader. The MAC layer then transfers the subframes, MAC subheader, and MAC header to the PHY layer. The PHY layer performs channel coding and modulation, and delivers the data to the receiver over the wireless channel afterward. An important aspect of this method is that instead of distributing the subheaders between the subframes, all the subheaders are concatenated and put in front of the subframes. The reason for such a design is that video streaming contains both data and control information, which should be treated with different priorities. Changing MCS over subframes is a common approach to support priority. However, when operating at a speed of gigabits per second, timely changing of the MCSs subframe by subframe can be difficult for the receiver. However, putting the subheaders in front enables the receiver to know the MCS of each subframe in advance, helping to realize timely MCS switching. It was reported that over 80 percent efficiency improvement and above 4 Gb/s throughput are achieved with standard aggregation [6].
In high-speed WPAN and WLAN systems, transmission efficiency decreases with the increase in transmission speed due to the increased ratio of overhead time to payload transmission time. To improve transmission efficiency and throughput performance, frame aggregation can be employed.
Low-Latency Aggregation — Low-latency aggregation is designed to support bidirectional communications for PC peripheral applications. These applications require aggregation to improve their transmission efficiency. However, they are sensitive to delay, because they need to frequently transmit very short command frames. Figure 5b shows the procedure of low-latency
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
119
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
The ultimate purpose of the 60-GHz WPAN systems is to deliver MAC throughput of the order of multi-Gb/s over a reasonable range. To accomplish this, system designers have to increase the transmission range, especially in non-line-of-sight channels.
aggregation. Unlike standard aggregation, the MSDUs from the upper layer are directly mapped into the subframes without fragmentation. The MAC subheader only contains global control information for all the subframes. The control information for individual subframes is distributed to each subframe as an MSDU subheader. Once there is one MSDU available, the transmission of the aggregated frame starts without waiting for the rest of the MSDUs to arrive. If no new MSDU arrives after all the available MSDUs are sent out, the MAC layer continues sending a special subframe, referred to as a zero-length MSDU, to fill the gap until new MSDUs become available. Compared with standard aggregation, transmission of the first part of the aggregated frame can start without waiting for the arrival of the whole frame, so the packaging delay of the aggregation process is reduced. However, some part of the control information has to be inserted between the subframes, which negatively impacts the efficiency. Assuming a frame payload of 512 bytes and a transmission speed of 1 Gb/s, a simulation found that up to 300 +s delay performance improvement can be achieved [8].
BEAMFORMING The ultimate purpose of 60 GHz WPAN systems is to deliver MAC throughput of multiple gigabits per second over a reasonable range. To accomplish this, system designers have to increase the transmission range, especially in NLOS channels [9]. To compensate for the high propagation loss in 60 GHz channels and reduce the effects of shadowing, the use of an antenna array has been proposed. The integration of multiple antennas into portable devices can be achieved easily, because the dimensions and necessary spacing of the 60 GHz antennas are on the order of millimeters [10]. As multiple antennas are available at both the transmitter and the receiver, multiple-input multiple-output (MIMO) techniques can be employed to increase spectral efficiency. However, MIMO techniques require multiple radio frequency (RF) chains, which significantly increase the complexity and cost. Therefore, the group focused on improving only the operational range with directional transmission based on antenna array beamforming. IEEE 802.15.3c specifies an optional beamcodebook-based beamforming protocol (BP) without the need for angle of departure, angle of arrival, or channel state information estimation. The proposed BP has the following features: • The BP consists of three stages: sector-level (coarse) searching, beam-level (fine) searching, and an optional tracking phase. The division of the stages facilitates a significant reduction in setup time compared with beamforming protocols with exhaustive searching mechanisms. • The BP employs only discrete phase shifts, which simplifies the DEVs’ structure compared to conventional beamforming with phase and amplitude adjustment. • The BP protocol is designed to be PHYindependent and is applicable to different antenna configurations.
120
Communications IEEE
A
BEMaGS F
• The BP is a complete MAC procedure. It efficiently sets up a directional communication link based on codebooks. Two types of BP are specified: on-demand beamforming and proactive beamforming. On-demand beamforming may be used between two DEVs or between the PNC and a DEV. It takes place in the CTA allocated to the DEV for the purpose of beamforming. Proactive beamforming may be used when the PNC is the source of the data to one or multiple DEVs. It allows multiple DEVs to train their own receiver antennas for optimal reception from the PNC, with lower overhead. During proactive beamforming, the sector-level training from PNC to DEV takes place in the beacon period. The sector-level training from DEV to PNC and the beam-level training in both directions take place in the CTAP. For both the beamforming modes, two forms of beamforming criteria are specified: beam switching and tracking (BST) criteria, and pattern estimation and tracking (PET) criteria. BST is suitable for any antenna configuration and must be supported by any DEV with beamforming capability. It is based on selecting the best beam from a given set of beams. On the other hand, PET is suitable only if 1D linear antenna arrays and 2D planar antenna arrays are employed in the DEV. Its support is optional, and is based on finding the optimal beamformer and beamcombiner vectors (i.e., antenna weights) that do not necessarily fall into the given set of beams.
CONCLUSION In this article we have explained 802.15.3c, which is the first IEEE standard achieving over 1 Gb/s at the MAC SAP. During its standardization, its task group not only created three new PHY modes, but also improved an existing MAC by adding aggregation and beamforming capability. Its channelization allows rapid deployment throughout the world. Already, customers can buy products based on the 802.15.3c standard. However, an open question left to the market to answer is which one of the PHY modes will be the most successful. Nevertheless, there are sufficient procedures within the standard to enable coexistence between the PHY modes. As a result, customers do not need to worry about interference from other compliant devices. From the standardization point of view, there are some possible features that were omitted due to the limited development time. For example, a PNC cannot enable multiple CTAs at a given time. In the future, a new task group could be created to add such features to improve the quality of enduser experience.
REFERENCES [1] FCC, “Amendment of Parts 2, 15, and 97 of the Commission’s Rules to Permit Use of Radio Frequencies Above 40 GHz for New Radio Applications,” FCC 95499, ET Docket no. 94-124, RM-8308, Dec. 1995. [2] IEEE Std 802.15.3c-2009 (Amendment to IEEE Std 802.15.3-2003), “IEEE Standard for Information Technology — Telecommunications and Information Exchange between Systems — Local and Metropolitan Area Networks — Specific Requirements. Part 15.3:
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for High Rate Wireless Personal Area Networks (WPANs) Amendment 2: Millimeter-Wave-Based Alternative Physical Layer Extension,” Oct. 2009, pp. c1–187. [3] A. Sadri, “Summary of the Usage Models for 802.15.3c,” IEEE 802.15-06-0369-09-003c, Nov. 2006. [4] A. A. M. Saleh and R. A. Valenzuela, “A Statistical Model for Indoor Multipath Propagation,” IEEE JSAC, vol. 5, Feb. 1987, pp. 128–37. [5] S.-K. Yong, “TG3c Channel Modeling Sub-committee Final Report,” IEEE 802.15-07-0584-01-003c, Mar. 2007. [6] S. Kato et al., “Single Carrier Transmission for MultiGigabit 60-ghz WPAN Systems,” IEEE JSAC, vol. 27, no. 8, Oct. 2009, pp. 1466–78. [7] F. Kojima et al., “Necessary Modifications on Conventional IEEE 802.15.3b MAC to Achieve IEEE 802.15.3c Millimeter Wave WPAN,” Proc. IEEE PIMRC 2007, Sept. 2007, pp. 1–5. [8] X. An et al., “Performance Analysis of the Frame Aggregation Mechanisms in IEEE 802.15.3c,” Proc. IEEE PIMRC 2009, Sept. 2009, pp. 2095–2100. [9] K. Sato et al., “60 GHz Applications and Propagation Characteristics,” IEEE 802.15-08-0651-01-003c, Sept. 2008. [10] J. A. Howarth et al., “Towards a 60 GHz Gigabit System-On-Chip,” Proc. 16th Wireless World Research Forum Meeting, Apr. 2006, pp. 1–8.
BIOGRAPHIES T UNCER B AYKAS (
[email protected]) __________ received his B.A.Sc degree from Bogazici University in 2000, and M.A.Sc. and Ph.D. degrees from the University of Ottawa in 2002 and 2007, respectively. He served as secretary and assistant editor for IEEE 802.15 WPAN Task Group 3c. Currently he is chair of IEEE 802.19 TG1, Wireless Coexistence in the TV White Space. He served as TPC vice chair of PIMRC 2009 and TPC chair of Cogcloud 2010. C HIN -S EAN S UM (
[email protected]) _________ received his M.E. from University Technology of Malaysia (UTM) in 2002 and his Ph.D. from Niigata University in 2007. In June 2007 he joined the NICT, Japan, as an expert researcher. He was involved in the IEEE 802.15.3c (TG3c) standardization activities, where he served as task group secretary and assistant editor. He is currently the coexistence contributing editor in IEEE 802.15.4g and an active contributor in IEEE 802.11af. Z HOU L AN (
[email protected]) _________ received his B.S. and Ph.D. degrees in electrical engineering from Beijing University of
Posts and Telecommunications (BUPT) in 2000 and 2005, respectively. He is currently with NICT of Japan as an expert researcher. He served as TPC vice chair of IEEE PIMRC 2009 and assistant editor of IEEE 802.15.3c. He is currently secretary of the IEEE 802.11af task group. He is also active in other IEEE WLAN and WPAN working groups.
IEEE
BEMaGS F
Already, customers can buy products based on the
JUNYI WANG (
[email protected]) ____________ received his B.Eng and M.Eng degrees from Harbin Institute of Technology, China, and his Ph.D. degree at Tokyo Institute of Technology. Currently he is an expert researcher at NICT working on TV white space communication systems. He is secretary of the IEEE P802.19 WG and TG1, and a voting member of the IEEE 802.15, IEEE 802.11, and IEEE 802.19 working groups. He was a guest editor of Wireless Communications and Mobile Computing.
802.15.3c standard.
M. A ZIZUR R AHMAN (
[email protected]) _________ received his Ph.D. from Niigata University in 2008. In recent years, while with NICT, he has worked on a number of projects focusing on millimeterwave, cognitive radio, wireless coexistence, white space communications, and sensing technologies, and contributed to wireless communications standardization within IEEE. He has authored/co-authored numerous articles and has over 25 patents pending. He was a recipient of the 2005 Japan Telecommunication Advancement Foundation Technology Award.
the most
However, an open question left to the market to answer is which one of the PHY modes will be successful.
HIROSHI HARADA (
[email protected]) __________ joined the Communications Research Laboratory, Ministry of Posts and Communications, in 1995 (currently NICT). Currently he is director of the Smart Wireless Laboratory at NICT and NICT’s Singapore Wireless Communication Laboratory. He is serving on the board of directors of the SDR Forum, as chair of IEEE DySPAN Standards Committee, and as vice chair of IEEE P1900.4 and IEEE 802.15.4g. He is a visiting professor at the University of Electro-Communications, Tokyo, Japan, and is the author of Simulation and Software Radio for Mobile Communications (Artech House, 2002). SHUZO KATO [F] (
[email protected]) _______________ currently is a professor, Research Institute of Electrical Communications, Tohoku University, Japan. He was program coordinator, Ubiquitous Mobile Communications at NICT working on wireless communications systems R&D. He served as vicechair of IEEE802.15.3c and chair of the Consortium of Millimeter Wave Systems for Practical Applications, promoting millimeter wave systems globally. He is a Fellow of IEICE Japan and served as an Editor of IEEE Transactions on Communications, Chairman of the Satellite and Space Communications Committee, IEEE ComSoc, and a Board Member of IEICE Japan.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
121
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
TOPICS IN STANDARDS
Overview of Femtocell Support in Advanced WiMAX Systems Ying Li, Samsung Telecommunications America Andreas Maeder and Linghang Fan, NEC Network Laboratories Europe Anshuman Nigam, Samsung India Software Operations Joey Chou, Intel Corporation
ABSTRACT The femtocell concept is an integral part of the telecommunication industry’s efforts to provide high-throughput, high quality services into the users’ home. In contrast to conventional cell types which are well-planned by the operators, femtocell base stations are supposed to be installed by customers themselves, similar to a WiFi access point. Unlike WiFi, however, femtocells operate mainly in licensed bands, such that operators are in control of the radio interface. This brings new challenges as well as opportunities for femtocell design; these include sophisticated mobility and interference management, increased reliability, as well as deployment in a plug-and-play manner. Extensive progress in femtocell design has been made in Advanced WiMAX recently, which is associated with the IEEE 802.16m update in 2011. This article gives an overview and update on novel concepts and mechanisms for femtocell support in the network architecture and air interface that have been adopted into the WiMAX Forum network specifications and the IEEE 802.16m specification.
INTRODUCTION The explosive increase in demand for wireless data traffic has created opportunities for new network architectures incorporating multitier base stations (BSs) with diverse sizes. Support for small-sized low-power BSs such as femtocell BSs is gaining momentum in cellular systems, because of their potential advantages such as low cost deployments, traffic offloading from macrocells, and the capability to deliver services to mobile stations which require large amounts of data [1–3]. Femtocells are supported by current thirdgeneration (3G) technologies and future nextgeneration cellular systems, such as Advanced WiMAX systems. Advanced WiMAX, which will provide up to 1 Gb/s peak throughput with the IEEE 802.16m [4] update in 2011, is one of the technologies for the ongoing International Mobile Telecommunications (IMT)-Advanced program [4, 5] for the fourth generation (4G) of mobile wireless broadband access. IEEE 802.16m defines the wireless metropolitan area
122
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
network (WirelessMAN) advanced air interface as an amendment to the ratified IEEE 802.162009 specification [6] with the purpose of enhancing performance such that IMT-Advanced requirements are fulfilled. In Advanced WiMAX, femtocell support is one of the solutions to provide high performance services even in indoor scenarios. In Advanced WiMAX, a femtocell BS, or WiMAX femtocell access point (WFAP), is a low-power BS intended to provide in-house and/or small office/home office (SOHO) coverage. With conventional macro- or microcells, indoor coverage is challenging and expensive to achieve due to the high penetration losses of most buildings. WFAPs are usually self-deployed by customers inside their premises, and connected to the radio access network (RAN) of the service provider via available broadband connections like digital subscriber line (DSL) or fiber to the home (FTTH). The self-deployment of WFAPs has implications on the requirements for operation and management. The WFAP must be able to react, in a highly flexible manner, to different interference situations, since neither the location nor the radio propagation environment can be predicted in advance. Furthermore, the customers are in physical control of the WFAP, meaning that they can switch it on or off at any time. Other factors like unreliable backhaul connections must also be considered. Considering these scenarios, some of the technical challenges and requirements can be identified as follows [7]: • Tight integration into the existing WiMAX architecture for support of seamless mobility between macrocells and WFAPs, lowcomplexity network synchronization, and localization • Advanced interference mitigation techniques to guarantee quality of service and coverage in macrocells as well as femtocells • Access control for different groups of subscribers as well as energy efficient-recognition of access rules by the mobile station (MS) • Support for increased reliability and autonomous reaction on irregular network or WFAP conditions
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
To design WFAPs capable of meeting these challenges and requirements, standardization efforts are being made in both IEEE 802.16 and WiMAX Forum. IEEE 802.16 Task Group m (TGm) defined the air interface support for femtocells. In parallel, the WiMAX Forum Network Working Group (NWG) is driving the development of specifications for femtocell solutions in the access and core networks. NWG defined two phases for femtocell support. In phase 1, support for the IEEE802.16-2009 air interface was specified in NWG Rel. 1.6 [8–10]. In phase 2, femtocell specifications to support the advanced air interface features as defined in IEEE 802.16m are developed. A high-level description of some of the technologies to support WFAPs can be found in [1, 2], but these have not yet been technically specified in the corresponding working groups at the time of writing. As IEEE 802.16m has just been completed, a significant number of technical details have been discussed, evaluated, and finally adopted into the specification. This article provides a detailed update on these recent developments in the standardization process for femtocell design in Advanced WiMAX. For example, no solution was discussed in [1, 2] for the coverage hole problem caused by the private femtocell interfering with the macrocell MS, but this article provides some solutions to it. For another example, this article also provides details on the final specification on femtocell reliability, femtocell low duty mode, access control, and interference management, which are not discussed in detail in [1, 2]. In the next section we describe how WFAPs are integrated into the WiMAX network architecture, which is mainly defined by the WiMAX Forum NWG. Advanced air interface support for mobility in WFAPs, interference mitigation, WFAP reliability, and WFAP low duty mode is introduced in later, mainly based on the most recent efforts in IEEE 802.16m. The article concludes with some summarizing observations and open issues.
NETWORK ARCHITECTURE FOR WIMAX FEMTOCELLS Figure 1 shows the general WiMAX network architecture with additional support for femtocells. In the following section, the main functional entities are described.
GENERAL WIMAX NETWORK ARCHITECTURE The network service provider (NSP) provides IP data services to WiMAX subscribers, while the network access provider (NAP) provides WiMAX radio access infrastructure to one or more WiMAX NSPs. A WiMAX operator may act as NSP and NAP. An NAP implements the infrastructure using one or more access service nodes (ASNs). An ASN is composed of one or more ASN gateways and one or more BSs to provide mobile Internet services to subscribers. The ASN gateway serves as the portal to an ASN by aggregating BS control plane and data plane traffic to be transferred to a connectivity service network (CSN). An ASN may be shared
by more than one CSN. A CSN may be deployed as part of a WiMAX NSP. A CSN may comprise the authentication, authorization, and accounting (AAA) entity and the home agent (HA) to provide a set of network functions (e.g. roaming, mobility, subscription profile, subscriber billing) that are needed to serve each WiMAX subscriber.
IEEE
BEMaGS
NETWORK ARCHITECTURE TO SUPPORT WIMAX FEMTOCELL For the support of femtocells, a Femto NAP and a Femto NSP are introduced. Additionally, SON (Self-Organizing Networks) functionalities are added. Femto NAP: A Femto NAP implements the infrastructure using one or more Femto ASNs to provide short range radio access services to femtocell subscribers. A Femto ASN is mainly differentiated from a Macro ASN in that the WFAP backhaul is transported over the public Internet. Therefore, the security gateway (GW) is needed for WFAP authentication. When a WFAP is booted, it first communicates with the bootstrap server to download the initial configuration information, including the IP address of the security GW. The WFAP and security GW authenticate each other, and create a secure Internet Protocol Security (IPSec) tunnel. The security GW then communicates with the femto AAA server in the femto NSP to verify whether the WFAP is authorized. The femto GW acts as the portal to a femto ASN that transfers both control and bearer data between MS and CSN, and control data between WFAP and femto NSP. Femto NSP: The femto NSP manages and controls entities in the femto ASN. The AAA function performs authentication and authorization of the WFAP. The management server implements management plane protocols and procedures to provide operation, administration, maintenance, and provisioning (OAM&P) functions to entities in the femto ASN. OAM&P enables the automation of network operations and business processes that are critical to WiMAX Femtocell deployment. WFAP management includes fault management, configuration management, performance management, and security management. SON functionalities to support femtocell: Self-optimized network (SON) functionalities can be divided into self-configuration and selfoptimization. Due to the large number of WFAPs expected, self-configuration is primarily intended to enable auto-configuration and avoid truck rolls. However, since femtocell deployments are not planned by operators, it is very important that the configuration (e.g., radio parameters setting) should take its neighbors into account by not adding interference to users. The wireless environment changes dynamically, as WFAPs can be powered on and off at any time. Self-optimization provides a mechanism to collect measurements from MSs and fine tune system parameters periodically in order to achieve optimal system capacity, coverage, and performance. Therefore, the SON server needs to interact with SON clients not only in the
F
The WFAP must be able to react, in a highly flexible manner, to different interference situations, since neither the location nor the radio propagation environment can be predicted in advance. Furthermore, the customers are in physical control of the WFAP.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
123
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Macro NAP
Macro ASN #N Macro ASN #1 CSN (NSP) BS
AAA
MS HA
ASN GW
BS
Femto NAP Femto ASN #N
Femto NSP
Femto ASN #1 Secure tunnel
Security GW
SON server
Management server
HA
AAA
Femto GW
WFAP
MS
Public Internet via DSL/cable Bootstrap server
WFAP operational states Power on/off Initialization
Operational state
- Network attachment - Synchronization - Topology acquisition
Normal operation mode
Control / management plane interface Data plane interface
Low duty mode
AAA: Authentication, authorization, and accounting ASN: Access service network CSN: Connectivity service network ASN GW: ASN gateway HA: Home agent NAP: Network access provider NSP: Network service provider Security GW: Security gateway SON: Self-organizing network WFAP: WiMAX femto access point BS: Base station MS: Mobile station
Figure 1. WiMAX network architecture.
femto ASN but also in the macro ASN. The information elements are exchanged between the SON server and the SON clients on the management plane. Therefore, they will be transported using the same management plane protocols as defined in the femtocell management specification. WFAP: For proper integration into the operator’s RAN, the WFAP enters the initialization state before becoming operational. In this state, it performs procedures such as attachment to the operators’ network, configuration of radio interface parameters, time/frequency synchronization, and network topology acquisition. After successfully completing initialization, the WFAP is integrated into the RAN and operates normally. In operational state, normal and low duty operation
124
Communications IEEE
modes are supported. In low duty mode, the WFAP reduces radio interface activity in order to reduce energy consumption and interference to neighboring cells. The low duty mode is discussed in more detail in a separate section.
AIR INTERFACE SUPPORT FOR WIMAX FEMTOCELLS The IEEE 802.16m standard amends the IEEE 802.16 WirelessMAN orthogonal frequency-division multiple access (OFDMA) specification to provide an advanced air interface for operation in licensed bands. It meets the cellular layer requirements of IMT-Advanced next-generation mobile networks and provides continuing sup-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
port for legacy WMAN-OFDMA equipment. In IEEE 802.16m, the basic features of the air interface are identical for femtocells and macrocells. For example, it supports both Time Division Duplex (TDD) and Frequency Division Duplex (FDD), it supports various bandwidth from 5 to 20 MHz for single carrier and wider bandwidth for multiple carrier, and it supports types of traffic such as VoIP, best- effort data, etc. However, there are some aspects for femtocells (as specified in Section 16.4 in IEEE 802.16m [4]) different from macrocells or other cells, mostly on Medium Access Control (MAC) layer and a few on Physical (PHY) layer. The IEEE 802.16m draft specification provides support for the operation of WFAPs and the integration of WFAPs in macrocell networks to provide functionality such as access control, network topology acquisition, mobility support, interference mitigation, reliability, and WFAP low duty mode. In the following section, these features are introduced in more technical detail. Discussions and contributed documents are archived in IEEE 802.16 Task Group m (TGm) website.
FEMTOCELL ACCESS CONTROL For the typical use case of WFAPs as a “private base station,” access control schemes must be supported. A closed subscriber group (CSG) containing a list of subscribers restricts access to WFAPs or certain service levels. IEEE 802.16m defines three modes for WFAP access; these are: CSG-Closed WFAPs are accessible exclusively to members of the CSG, except for emergency services. CSG-Open or hybrid WFAPs grant CSG members preferential access. However, subscribers which are not listed can still access the WFAP at a lower priority. OSG (Open Subscriber Group) WFAPs are accessible by any MS much like a normal macro BS. For efficient identification of subscriptions and accessibility of WFAPs, a femto-enabled MS can maintain a CSG whitelist, containing a set of WFAPs and corresponding attributes like geographical location or overlaying macrocell identifiers. To avoid large CSG whitelists, a CSG identifier (CSGID) is defined which describes a group of WFAPs within the same CSG. The CSGID can be derived directly without any additional information from the global unique BS identifier (BSID) of the WFAP.
NETWORK TOPOLOGY ACQUISITION Knowledge of the network topology is critical for efficient interference mitigation and mobility management between macrocells and WFAPs and among WFAPs. Both macrocells and WFAPs have to be aware if a WFAP enters or leaves the environment, thus changing interference and mobility conditions. Furthermore, MSs can perform cell searching and handovers in a more efficient way if the type and the access policies of the WFAPs in connection range are known beforehand. IEEE 802.16m supports MSassisted network topology acquisition, but the WFAPs can also scan the radio environment to
find neighbor or overlay cells. Figure 2 shows some approaches for network topology acquisition.
IEEE
BEMaGS
MS Acquisition of WFAP Topology — IEEE 802.16m adopts an energy-efficient two-step scanning method for the MS to identify neighboring WFAPs, and further to efficiently identify whether an MS is allowed to access the WFAP. Identified WFAPs and their attributes can then be reported to overlaying macrocells and neighboring WFAPs. Base station types are differentiated by the frame preamble sequence, which is uniquely mapped to an IDcell identifier. The total number of preamble sequences is partitioned into subsets to differentiate between BS types. To make the scanning and possible network entry efficient, the set of IDcells is partitioned into sets for macro and non-macro cells, where the latter set is further partitioned into private (further partitioned into CSG-closed and CSG-open) and public cells (further partitioned into pico and relay). The two-step scanning method works as follows. In the first step, an MS scans the frame preamble sequence to determine the BS type. However, the number of WFAPs within the coverage area of a macrocell may well exceed the number of available IDcell identifiers, such that the identity of a WFAP may not be resolved uniquely. To solve this WFAP ambiguity problem, the MS decodes in the second step the periodically broadcasted superframe header (SFH) to obtain the unique BSID identifier. Note that to save battery energy, the second step is only performed if necessarily. In the second step, the MS can also derive the CSGID of the WFAP, and compare with its local CSG whitelist to determine whether the detected WFAP is an accessible cell for the MS.
F
The IEEE 802.16m draft specification provides support for the operation of WFAPs and the integration of WFAPs in macrocell networks to provide functionality such as access control, network topology acquisition, mobility support, interference mitigation, reliability, and WFAP low duty mode.
WFAP Acquisition of Neighboring Cells — The WFAP can acquire the network topology from the backhaul, from the reporting MSs, or by active scanning. The WFAP can scan and measure its neighboring cells, such as overlaying macrocells, or other nearby WFAPs, for interference management and to assist the cell (re)selection of the MS. The WFAP in this way acts like an MS. However, in TDD (time division duplex) systems, the WFAP cannot transmit frame preambles and SFH during scanning. Hence, the WFAP broadcasts a SON-ADV (SON advertisement) message which includes the timing information of the scanning interval, in which the WFAP scans the other cells, while its own preambles and SFH may not be available for the MS in its coverage to scan. This message informs MS that WFAP is not available for scanning.
FEMTOCELL MOBILITY MANAGEMENT Femtocell networks, especially in the case of dense deployments, are challenging for mobility and hand-over functions due to the large number of small cells with different access types. Special focus must be set on cell scanning functions to avoid high energy consumption on the MS side. Also, seamless handover must be supported to avoid QoS degradations. Figure 3
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
125
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
MS scanning neighboring BSs
MS scanning neighboring WFAPs
Level one scanning MS containing information on IDcell partitioning for different cell types
SON-ADV message: containing the timing of the scanning interval in which 1) the WFAP scans its neighboring cells, 2) unavailable for the MS.
Synchronization channel, containing preamble, or IDcell MS MS determines whether the scanned BS is a CSG-closed, CSG-open, Open type.
Macro BS
WFAP
WFAP
H, et c.
Level two scanning is less frequent than level one scanning
Pr
Macro BS
ea m bl e, SF
Preamble, SFH, etc.
n. atio nic eport t u m yr ul ComS ma g res M nnin sca
MS
Level two scanning MS containing white list: a list of the femto CSG the MS subscribes
SON-ADV: unavailable interval
SFH, containing BSID, from which CSGID can be derived. Zero overhead for CSGID. WFAP
MS WFAP
n. atio nic eport t u m yr ul ComS ma g res M nnin sca
MS gets femto BSID and CSGID. If the received CSGID is in its local white list, the MS is a subscriber of the CSG.
WFAP uses an interval to scan its neighboring cells
Macro BS
Figure 2. Network topology acquisition.
shows some optimized mobility management support in IEEE 802.16m. Optimized MS Scanning of WFAPs — Macrocell BSs and WFAPs can help MSs in the process of scanning for WFAPs by conveying information on the WFAP network topology. This is achieved by broadcast, unicast, and request-response message exchanges. Specifically, a macrocell BS can broadcast information on OSG WFAPs in their coverage area like carrier frequencies or IDcell partitions to reduce the scanning time for MSs. Furthermore, after successful association of an MS to the macrocell network, a macrocell BS can transmit a list of accessible neighboring WFAPs. An MS may also explicitly request a list of accessible WFAPs. An MS may request additional scanning opportunities from a BS by sending a message including the detected IDcell index and carrier frequency information. Upon reception of the message, the BS can respond with list of accessible neighbor WFAPs. Scanning of closed-subscriber group WFAPs should be minimized as far as possible as long as the subscriber is not authorized. Therefore, the MS may provide CSGIDs of CSG whitelists to the current serving base station to obtain instruction on how and when to scan, these instructions may include a list of WFAP BSIDs which are associated with the requested CSGIDs. Furthermore, information on the location of
126
Communications IEEE
WFAPs is also exploited to optimize MS scanning of WFAPs (e.g., by triggering MS scanning when the MS or the network judges that the MS is near a WFAP) based on the location of the WFAPs and MS. The CSG whitelist may include location information of CSG WFAPs, such as GPS info or overlay macrocell BSID. The network may also instruct (by sending a message, which may include a list of allowable WFAPs nearby) the MS to scan WFAPs based on location information available at the network. Handover Support for WFAPs — Handovers between macrocells and WFAPs as well as interWFAP handovers should be transparent and seamless for high QoS, besides taking user preferences into account. For example, subscribers may prefer their home WFAPs even if the signal strength of the WFAPs is lower than that of adjacent macrocell BSs. To this end, the WMAN-advanced air interface defines handover and scanning trigger conditions, and target BS priorities for femtocells based on the BS type. Trigger conditions can be defined such that an unwanted handover from a home WFAP to the macrocell network is avoided, or vice versa handovers to WFAPs are preferred. In addition, the network can instruct the MS on how to prioritize the cell (re)selection. For example, the network or the serving BS can send the MS a message that includes a prioritized list of the candidate target base stations.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Since WFAPs are
MS can scan autonomously or based on the instruction by BS.
MS can have preferences on its subscribed femto
F
MS and its serving BS negotiate the timings for MS to scan its neighboring WFAPs. MS can report its location, detected BS, subscribed CSGIDs, etc.
MS
overlaid by macroServing macro BS
Serving BS can unicast an optimized neighbor list which includes possibly accessible WFAPs, to help the MS scan neighboring WFAPs energy-efficiently
cells, interference occurs not only in the tier of WFAPs, but also across tiers, i.e. between WFAPs and macrocells. Advanced interfer-
Targeting WFAP
Note: Different types of BS can apply different trigger conditions for handover
ence management is therefore crucial for viable operation of femtocell networks, especially in
Figure 3. Optimized handover scheme.
dense-deployment
INTERFERENCE MANAGEMENT Since WFAPs are overlaid by macrocells, interference occurs not only in the tier of WFAPs, but also across tiers, i.e. between WFAPs and macrocells. Advanced interference management is therefore crucial for viable operation of femtocell networks, especially in dense-deployment scenarios. Several factors need to be considered, such as whether the WFAP and overlay macrocell use the same frequency, whether multiple frequency carriers are available, whether interference management is applied to control channel or data channel or both, and whether the interfered MS is in connected or in idle mode. In order to provide seamless connectivity and high QoS to mobile stations, the WMANadvanced air interface supports advanced interference management methods with a set of technologies targeting different scenarios. The purpose is to achieve efficient inter-cell interference mitigation with acceptable complexity in an optimized manner. Advanced interference management crosses multiple layers, such as physical layer (e.g., power control, carrier change), MAC layer (e.g., signaling, messages, resource management such as resource reservation), network layer (e.g., security, SON server coordination), and other higher layers (e.g. mobile station QoS requirement and provisioning). Some of these technologies are described below: Resource Reservation and Blocking — A CSG WFAP may become a strong source of interference for non-member MSs which are associated to a nearby macrocell. In this case, the WFAP blocks a radio resource region (i.e. a time/frequency partition of the radio frame) exclusively for non-member MSs for communication with the macrocell BS. This approach may reduce the capacity of the WFAP. However, since typically the number of MSs served by WFAP is expected to be low, in many cases it is suitable and it does not severely downgrade subscribers’ services.
Power Control — A CSG WFAP adjusts the transmit power to reduce interference at nonmember MSs. For example, the transmit power may be reduced to satisfy the minimum QoS requirements of its member MSs if the WFAP is strongly interfering non-member MS(s). The power level may be restored again to provide better QoS to its member MSs as soon as the non-member MS left the coverage area of the WFAP. This approach may reduce the WFAP coverage or it may reduce the subscribers’ throughput. It can be suitable for indoor scenarios with high wall penetration losses between macro and WFAPs, or in scenario where no MSs are served by WFAP at the edge, or the subscribers’ throughput requirement is not high.
scenarios.
Coordinated Fractional Frequency Reuse — Both WFAP and macrocells coordinate their frequency partitions and the associated power levels over the frequency partitions. This approach can be suitable for interference management of user data channels, but it may not help the control channel interference management. In addition, the cell may have reduced throughput due to resources restrictions. Frequency Carrier Change — The WFAP can change to another frequency carrier with less interference if there are more than one frequency carriers available. This approach requires more than one frequency carriers. Spatial Coordinated Beamforming — If beamforming is supported by the WFAP, the WFAP and/or the macrocell can coordinate their antenna precoding weights to avoid or mitigate interference. This approach requires timely signaling between WFAP and macrocell. The coordination may need accurate cell synchronization which can be challenging. Femtocell-Macrocell Coordinated Handoff Scheme — A CSG WFAP can hand off some of its member MSs to a nearby macrocell so that the WFAP can adjust radio resources (e.g., by means of power control or radio resource reser-
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
127
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Note that IEEE 802.16m specifies different techniques
A
BEMaGS F
SON server, etc.
Can coordinate with cells, e.g., macro BSs and WFAPs, optimizing configuration and parameters, such as transmitting power, carrier frequency, reserved resource, WFAP access type, etc.
Backhaul
as enablers for interference manage-
Step 2: WFAP negotiates with network and performs interference mitigation
ment. The implementation and utilization of these techniques is not mandatory, such that operators and vendors can choose
Macro BS Member MS Step 1: MS signals to WFAP about its situation
which set of options
Non-member MS
is most suitable for their concrete deployment scenario.
WFAP
Macrocell MS, who is not a member of CSGclosed WFAP. The MS cannot connect with macro BS due to the interference from WFAP.
Figure 4. Two-step interference management in case a CSG-closed WFAP generates high interference at a non-member MS.
vation) to reduce interference to non-member MSs served by the macrocell. The timing of the resource adjustment can be adaptively set to accommodate the QoS requirements from both WFAP member MSs and non-member MSs. This approach may bring an accounting disadvantage to the member MSs if the communication with WFAP is cheaper than with the macro, and it may weaken the WFAP to offload the traffic in macrocell. Femtocell Type Change Under Service Agreement — If required, a CSG WFAP can temporarily change its subscriber type (e.g., from CSG-Closed to CSG-Open) if it strongly interferes with a non-member MS, such that the MS can hand over to the now CSG-Open WFAP. The subscriber type is restored as soon as the non-member MS leaves the coverage area of the WFAP. This approach requires that the owner of the WFAP agrees that other users temporarily use the WFAP for data services. One of the biggest problems for the operation of heterogeneous macrocell/WFAP deployments is the creation of coverage holes for macrocell users by CSG-closed WFAPs. If a mobile station is not a member of the subscriber group of a CSG-closed WFAP, the received signal power is experienced as interference. This may lead to service degradation and in the worst case to connection loss — i.e. a coverage hole is created. To solve this problem, IEEE 802.16m defines a two-step solution, as shown in Fig. 4. Step 1: After scanning, an MS detects that the only BS with acceptable signal quality is a CSG-closed WFAP where the MS is not listed as member. Normally, a non-member MS should not try to access the CSG-closed WFAP [3, 7]. However in the exceptional case of a coverage hole generated by the CSG-closed WFAP, the non-member MS can signal the coverage hole
128
Communications IEEE
situation to the WFAP by means of a reserved CDMA ranging code. Step 2: The WFAP can notify the macrocell and a network entity such as a SON server to request coordinated interference mitigation. It has to be noted that the coordinated interference management not only means that the nonmember MS served by the macrocell BS will get desired QoS, but also the WFAP tries to guarantee its member MSs desired QoS. Depending on the scenario, interference mitigation approaches such as resource reservation, power control, FFR, or beam-forming can be applied. The two-step approach can be used in a general case when an MS is connected with a macrocell, where in step 1 the MS can report the interference to the macro BS, and in step 2 the macro BS can coordinate with the interfering WFAP via the backhaul network and then the interference mitigation approaches can be applied. Note that IEEE 802.16m specifies different techniques (see examples above) as enablers for interference management. The implementation and utilization of these techniques is not mandatory, so operators and vendors can choose which set of options is most suitable for their concrete deployment scenario. How to make choices would be open to operators and vendors, and further simulation studies may be needed to make the choice.
WFAP SERVICE RELIABILITY Since WFAP BSs are under physical control of the customers, normal operation may be interrupted for various reasons. Typical examples are loss of power support or backhaul connectivity problems. Also, operators may schedule maintenance times and network topology reacquisition or interference mitigation procedures. However, service continuity should be maintained as much
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
as possible in these cases. IEEE 802.16m introduces features for increased reliability and continuous service to the MS (Fig. 5) should such scenarios arise. As a basic rule, the WFAP will try to inform the network and MSs in case of any service disruptions. On the air interface, this is done by means of a periodically broadcasted message. The message encodes the reason for the unavailable time, relevant system parameters like transmit power reduction and frequency allocation index, and eventually the duration of the air interface absence, if known beforehand. Additionally, a list of recommended BSs for the MS to handover can be included. The message is broadcasted until the WFAP disables the air interface. This allows the MS to initiate a handover to a BS based on the recommended list or to any previously cached neighbor BS list of its preference. Alternatively, the WFAP can instruct the MS to handover to other BSs before a scheduled unavailable time is due. For optimized network re-entry to a WFAP that becomes available again, the WFAP may store medium access control (MAC) context information of the served MSs (e.g. basic capabilities, security capabilities, etc.). If the WFAP recovers from failure of backhaul, or power down, or reconfiguration or it regains some resources from interference coordination, it may inform the network or notify the current serving BS of the MS through the backhaul network interface. Based on the cell types of the current serving BS and the WFAP and the associated mobility management policy, the current serving BS may then initiate a handover back to the WFAP, where the recovered WFAP is prioritized.
WFAP OPERATION IN LOW DUTY MODE A novel optional operational mode, low duty mode (LDM), was introduced into IEEE 802.16m in order to reduce interference and energy consumption in femtocell deployments. The principle of the LDM is to reduce air interface activity as much as possible by transmitting on the air interface only if it is required. To this end, default LDM patterns consisting of available intervals (AI) and unavailable intervals (UAI) are defined which enables a pattern of activity and inactivity for the WFAP. The UAI allows disabling or switching off of certain parts of WFAP hardware components such as the transmitter chain. Another possibility is to use UAIs for scanning and measurement of the radio environment in order to improve interference mitigation or for synchronization to the macrocell network. During an AI, the WFAP is available for any kind of transmission just as in normal operation state besides being guaranteed to be available for scanning by the AMSs. The LDM is designed with two basic paradigms: First, a WFAP may enter LDM only in case there is no active MS attached. This rule is established in order to avoid complex signaling and possible QoS degradation at the user side. Second, the impact on the operational complexity of the MS should be minimized in order to keep implementation costs low. The Default LDM pattern is either pre-provi-
IEEE
BEMaGS F
SON-ADV message: self-organizednetwork advertisement message, including the reason of WFAP being unavailable, reconfiguration information, etc. Macro BS MS
Handover to macro
SON-ADV message Power WFAP Cable WFAP reduces or reconfigures its air interface resource for interference management, etc.
Figure 5. Illustration of WFAP reliability design. sioned, unicasted during network entry or may be broadcasted, such that MSs have the necessary information on when the WFAP is available, for example for requesting bandwidth. The WFAP switches back to normal mode on explicit request from the backhaul network or implicitly by receiving any triggering of data activity from the MSs. An AI will be scheduled by the WFAP whenever there is an operational need for it. Therefore, the resulting AI pattern at the WFAP is the superposition of the Default LDM pattern and any additional AIs necessary for normal MS operation. This is illustrated in Fig. 6. Note that for interference mitigation, it is desirable to reduce the transmitting time of the LDM as much as possible, for example by aligning paging and LDM Default Patterns.
SUMMARY The femtocell concept is supported in Advanced WiMAX for low-cost deployment, high throughput, and high quality of service in indoor scenarios. Recent standardization activities in IEEE 802.16m and the WiMAX Forum, the technical details for the next generation WiMAX femtocell design have been defined. Advanced WiMAX introduces innovative solutions for femtocell support into the WiMAX network architecture and the WirelessMAN-Advanced Air Interface. This article highlights the challenges and design principles of Advanced WiMAX Femtocells. The features and mechanisms that solve the unique problems of deploying and operating femtocell networks have been illustrated. These include network topology acquisition, enhanced mobility management, coordinated interference management, increased service reliability, and operation of low duty mode. Some of the technologies to support femtocells in Advanced WiMAX may be further researched and optimized, such as coordinated interference management, operation of low duty mode, and enhanced mobility management. System performance evaluation, modeling, and simulations may be further studied. Furthermore,
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
129
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
MS paging cycle
Some of the technologies to support femtocells in Advanced WiMAX may be further researched and optimized, such as coor-
PLI
WFAP transmission pattern
Default LDM pattern
ment, etc. System performance evalua-
MS paging cycle
WFAP default LDM pattern
management, opera-
mobility manage-
F
MS in idle mode
AI
mode, enhanced
BEMaGS
PUI
dinated interference tion of low duty
A
UAI
AI
UAI
AI UAI
Default LDM pattern
AI
UAI
UAI
AI
Default LDM pattern
AI: Available interval UAI: Unavailable interval PLI: Paging listening interval PUI: Paging unavailable interval
Figure 6. WFAP operation in low duty mode.
tion, modeling and simulations, may be further studied.
some topics are not included in the IEEE 802.16m air interface, such as support for downlink transmission power setting or power control of femtocells and multicarrier support in femtocells. These topics may be included in study items in future standards releases. In May 2010 the study group on Hierarchical Networks started and is ongoing since, one of whose aims is to extend the IEEE 802.16m air interface for multitier deployments. A formal task group on this topic is expected to be created.
REFERENCES [1] S.-P. Yeh et al., “WiMAX Femtocells: A Perspective on Network Architecture, Capacity, and Coverage,” IEEE Commun. Mag., vol. 46, no. 10, Oct. 2008 [2] R. Y. Kim, J. Sam Kwak, and K. Etemad, “WiMAX Femtocell: Requirements, Challenges, and Solutions,” IEEE Commun. Mag., vol. 47, no. 9, Sept., 2009. [3] D. N. Knisely, T. Yoshizawa, and F. Favichia, “Standardization of Femtocells in 3GPP,” IEEE Commun. Mag., vol. 47, no. 9, Sept., 2009. [4] IEEE Std 802.16m-2011, “Part 16: Air Interface for Broadband Wireless Access Systems Amendment 3: Advanced Aire Interface,” May 2011. [5] ITU-R, “IMT-ADV/8-E, Acknowledgement of Candidate Submission from 3GPP Proponent (3GPP Organization Partners of ARIB, ATIS, CCSA, ETSI, TTA AND TTC) under Step 3 of the IMT-Advanced Process (3GPP Technology),” Oct., 2009. [6] IEEE 802.16-2009, “IEEE 802 Part 16: Air Interface for Broadband Wireless Access Systems.” [7] IEEE, “IEEE 802.16m-07/002r10, IEEE 802.16m System Requirements,” Jan. 2010. [8] WiMAX Forum, “DRAFT-T33-118-R016v01-C_FemtoCore,” May 2010. [9] WiMAX Forum, “DRAFT-T33-119-R016v01-D_Femtomgmt,” May 2010. [10] WiMAX Forum, “DRAFT-T33-120-R016v01-C_SON,” June 2010.
BIOGRAPHIES ____________ received her Ph.D. degree YING LI (
[email protected]) in electrical engineering from Princeton University, New Jersey, in October 2008. She received her B.E. degree (with honors) in electrical engineering from Xi’an Jiaotong University, China. Since October 2008 she has been with Samsung Telecommunications America, Dallas, Texas, where she is involved in study, development, and standardization of heterogeneous networks for next - wireless communications. She is actively involved in IEEE 802.16m standardization, especially for femtocell support, in which she has chaired ad -hoc sessions in the IEEE 802.16m technical working group. Her current research interests include wire-
130
Communications IEEE
less networks, heterogeneous networks, resource allocation, cross-layer design, and optimization. ANDREAS MAEDER (
[email protected]) _______________ received his diploma and doctoral degree in computer science from the Department of Distributed Systems at the University of Würzburg, Germany, in 2003 and 2008, respectively. Since 2008 he has been affiliated as a research scientist with NEC’s Network Laboratories Europe in the Mobile and Wireless Network Group. His current responsibilities include IEEE 802.16m standardization with a special focus on femtocells, IMT-Advanced evaluation, and research on radio resource algorithms. He was also involved in the development of femtocell solutions for 3GPP LTE systems. His research interests focus on radio resource management schemes for IMT-Advanced systems, femtocells, and M2M communications. LINGHANG FAN (
[email protected]) _______________ is a consultant with NEC Laboratories Europe. He received his B.Eng. in automatic control from Southeast University, China, and his M.Sc. and Ph.D. in telecommunications from the University of Bradford, United Kingdom. In 2003 he joined the University of Surrey as a postdoc research fellow and worked on the EU projects STRIKE, Ambient Networks, MAESTRO, SatNEx, SATSIX, and EC-GIN. His research interests include wireless/mobile communications and next-generation Internet. From 1998 to 2000 he was a researcher at the University of Bradford and worked on the EU projects SINUS and SUMO. He has published more than 50 papers in international journals and conferences, and edited a book. A NSHUMAN N IGAM (
[email protected]) _____________ received his Bachelor’s degree in electrical engineering from the Indian Institute of Technology at Kanpur in 2000. Since 2000 he has been affiliated with Samsung India. He has worked on the design and development of the access part of GPRS, UMTS, LTE, and WiMAX systems. His current responsibilities include IEEE802.16m standardization where he is actively participating in the femtocell and control plane areas. He is also actively involved in design and development of IEEE802.16m systems. His research interests are in femtocells and cooperative communications in wireless mobile networks. J OEY C HOU (
[email protected]) ____________ received M.S. and B.S. degrees in electrical engineering from Georgia Institute of Technology and the National Taiwan Institute of Technology, respectively. Since 2003 he has been working in the creation and evangelization of WiMAX technology in IEEE 802.16 and WiMAX Forum. He worked on the product development of the first WiMAX chipset, Rosedale at Intel. He played several key roles in the WiMAX standardization in both IEEE 802.16 and WiMAX Forum, including chief technical editor and team lead. He was actively involved in standard developments in the ATM Forum and DSL Forum. Prior to joining Intel, he worked at Siemens, AT&T, and Motorola, where he worked on the Iridium and Teledesic projects.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
____________________
_________________________________________
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
TOPICS IN STANDARDS
Smart Utility Networks in TV White Space Chin-Sean Sum, Hiroshi Harada, Fumihide Kojima, Zhou Lan, and Ryuhei Funada, NICT
ABSTRACT This article presents an overview of the background, technology, regulation, and standardization in the course of deploying smart utility networks (SUNs) in TV white space (TVWS) communications, two wireless technologies currently receiving overwhelming interest in the wireless industry and academia. They are independent to and uncorrelated with each other, but share the same mission: conserving resources and increasing efficiency. This article reviews the systems as separate technologies, and then combines them to propose a hybrid solution that draws out their respective advantages. The first part focuses on SUNs and describes the SUN usage model with typical application requirements and practical examples, followed by the latest developments in standardization initiatives with emphasis on the currently active IEEE 802.15.4g and 802.11ah groups. The second part discusses TVWS, studying and summarizing the regulations governing its usage, and then reports on the standardization bodies’ responses to these regulations with a focus on IEEE 802.11af, 802.19.1, and 802.22. Finally, the third part amalgamates the SUN usage model with TVWS regulations and deployment scenarios, providing relationship mapping between the SUN components and regulation-compliant TVWS devices. Further discussions concentrate on the opportunities and challenges along the path of realizing a practical SUN in the TVWS spectrum under the current technical and regulatory conditions. Several recommendations are made from both regulatory and technical standpoints to further increase utilization of SUNs in TVWS.
INTRODUCTION It is the best of times, it is the worst of times. Scarce radio resources have arrived at a bottleneck wherein communication systems must compete and struggle in order to deliver adequate performance. Meanwhile, demand for “ecofriendly” radios has also opened a new paradigm and direction for radio designers to mobilize their efforts. Communication systems that are able to conserve or reuse the already scarce radio resources are currently leading the way in next-generation
132
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
radio design. While minimizing potential impacts on the congested radio spectrum, next-generation communication systems are expected to help manufacturers and operators achieve longterm profitability. Mobilizing efforts to spur progress of these eco-friendly systems has recently become not only the mission of communication engineers, but also that of every resident feeding on the Earth’s resources. In line with these green-related efforts, the main intent here is to combine two eco-friendly systems to produce a hybrid technology. We hope that this innovation can bring more effective use of radio resources while allowing a broader market to be explored. This article’s main objective is to provide a high-level discussion to explore the potential of these technologies in changing our lifestyles, based on the current technology, market and regulatory status. The first eco-friendly system of interest is the smart utility network (SUN). A SUN is a telemetry system closely related to the smart grid framework, which targets designing a modernized electricity network as a way of addressing energy independence, global warming, and emergency response. A SUN is a ubiquitous network that facilitates efficient management of utilities such as electricity, water, natural gas, and sewage. Effective management of utility services is expected to encourage energy conservation and reduce resource wastage. The second system is TV white space (TVWS) communication, which occupies the frequencies allocated to TV broadcasting services but not used locally. In other words, it is an enabling technology to reuse the spectrum not used by primary licensed users. With the spectrum below 10 GHz becoming more congested, this is a timely opportunity for secondary wireless systems to access the surplus TVWS. This article promotes combining both technologies into a hybrid eco-friendly system. By deploying the SUN telemetry system in TVWS bands, we are able to obtain both the energy conservation of a SUN and the spectrum reusability of TVWS, thus creating a hybrid ecofriendly wireless technology. Both systems are now popular topics in industry and academia, and it is only a matter of time until a hybrid SUN-in-TVWS wireless system becomes the focus of attention. It is also notable that while
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
F
Data collectors form a mesh network and are connected to the External network
utility providers via Utility provider 2
wireless or other wired solutions. In the case of an
Data collector
Utility provider 1
emergency or link outage of fixed data collectors, a mobile data collector may
Data collector
be deployed as an alternative. Data collector
Electric meter
SUN link
Gas meter Mobile data collector
Water meter
Figure 1. Usage model of smart utility networks.
amalgamating SUN with TVWS brings encouraging opportunities, there are several accompanying challenges. To address the issues at hand, the article is organized as follows. We discuss the usage model of the SUN and the international standardization initiatives working on related topics. We also discuss the regulations, deployment scenarios, and corresponding standardization initiatives of TVWS communications. Then we analyze the potential of SUN occupying TVWS, with weighted focus on the opportunities and challenges therein. Finally, we provide several recommendations on plowing through the challenges to harvest the advantages of a hybrid SUN-inTVWS system.
SMART UTILITY NETWORKS Smart utility networks (SUNs) are next-generation utility networks allowing digital technologies to provide efficient control and management of utilities such as electricity, water, natural gas, and sewage. One key technology of SUNs is the advanced metering infrastructure (AMI), which facilitates monitoring, command, and control for service providers at one end, and measurement, data collection, and analysis for consumers at the other. AMI covers a wide range of technologies including communication protocols, hardware, software, data management systems, and customer-related systems. Communication technology is the essential element for enabling for-
mation of networks where control messages and metering data can be exchanged. This section discusses the potential usage scenario of SUNs, and gives an overview of recent developments in international standardization initiatives.
SUN USAGE MODEL Conventional utility control and metering are typically performed by manual or semi-manual operations with many limitations. For utility service providers, the time is right to conduct a paradigm shift to a more intelligent networking system to increase service efficiency and costeffectiveness. Radio frequency (RF)-based mesh networking systems that improve the quality of conventional utility networks provide a good usage model with huge market potential. To realize such an efficient and inexpensive utility network, several key application properties are needed. First, the network must be ubiquitous and far-reaching. Since there are potentially millions of nodes to be deployed, with multiple nodes per customer, it is crucial to have “everyone connected to everyone” to ensure continuous connectivity via self-healing and self-forming networks. The network must also be responsive to system failure. In emergencies such as service outages, real-time response is a critical factor for recovery. Third, the network must be scalable. Given the diverse population density from metropolitan areas to sparse rural towns, the network must be able to support a range of pervasive traffic loads. Lastly, the net-
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
133
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
To align with the environmental conditions encountered in smart metering deployments, a SUN is required to achieve an optimal energyefficient link margin. It is also required to
Regulatory domain
Frequency band
Worldwide
2400–2483.5 MHz
Japan
950–958 MHz
United States
902–928 MHz
Europe
863–870 MHz
China
470–510 MHz
Table 1. Major allocated frequency bands for SUN.
support up to three collocated orthogonal networks.
work must be cost effective. Network components should be easily deployed, with low complexity and a long life span. Figure 1 illustrates the RF-based utility mesh network known as the SUN. In the usage model, meters of different utilities, such as in a residence, are connected via the SUN link. Each house is connected to another, also via the SUN link. A collective number of households form a service area that is covered by a data collector. Data collectors form a mesh network and are connected to the utility providers via wireless or other wired solutions. In the case of an emergency or link outage of fixed data collectors, a mobile data collector may be deployed as an alternative. Figure 1 gives an overview of the primary usage case for the SUN. The usage model in the figure should also support a wide range of applications. Apart from the most commonly known, automatic meter reading (AMR), these applications include remote service connect/disconnect, outage detection, reliability monitoring, and quality monitoring.
DEVELOPMENT IN STANDARDIZATION INITIATIVES The usage model mentioned earlier is seen as a good scenario for next-generation utility networks. This section presents recent developments in some of the latest standardization initiatives targeting specifications of these networks. IEEE 802.15.4g — IEEE 802.15.4 is a set of standards specifying communication systems in wireless personal area networks (WPANs). The standards are collectively managed by a mother entity, the 802.15 Working Group (WG). The WG distributes specific task areas to smaller subgroups, called Task Groups (TGs), and each TG is assigned to standardize a particular field in the WPAN system. For the next-generation utility system, IEEE 802.15.4g, better known as TG4g, is responsible for developing the SUN standard. TG4g specifies the necessary SUN-related physical (PHY) layer design amendments to legacy IEEE 802.15.4 [1], the base standard for low-data-rate WPANs. Medium access control (MAC) amendments on legacy IEEE 802.15.4 [1], however, are handled by a separate TG
134
Communications IEEE
A
BEMaGS F
called IEEE 802.15.4e, or TG4e. The MAC amendments specified by TG4e provide the MAC to be applied by TG4g SUN operations. The TG4g scope is for specifying a standard that supports SUN operability in regionally available unlicensed frequency bands, as shown in Table 1. The table shows just a fraction of the complete list of frequency bands listed in TG4g. Targeted data rates range from 40 to 1000 kb/s corresponding to different deployment scenarios and traffic conditions, principally in outdoor communications. To align with the environmental conditions encountered in smart metering deployments, SUN is required to achieve an optimal energy-efficient link margin. It is also required to support up to three co-located orthogonal networks. A typical deployment scenario is connectivity to at least 1,000 direct neighbors in a dense urban network. In TG4g PHY layer design, three alternative PHY layer specifications are proposed: multirate frequency shift keying (MR-FSK) PHY, multirate orthogonal frequency-division multiplexing (MR-OFDM) PHY, and multirate offset quadrature phase shift keying (MR-O-QPSK) PHY. The multiple PHY designs are for tackling different system demands in respective market segments. The PHY frame size is set to be a minimum of 1500 octets. To promote coexistence, a multi-PHY management (MPM) scheme is proposed to bridge the three PHYs using a common signaling mode (CSM). For MAC layer design, TG4g principally reuses the MAC protocols in [1], which employs a PAN maintained by the PAN coordinator (PANC). The PANC manages and controls the time and spectrum resources to be shared among the devices within the PAN. A device capable of starting its own PAN is allowed to do so within an existing PAN, thus enabling a hierarchical cluster tree type of network. Most of the MAC protocols and functionalities in [1] are preserved in the MAC design for SUN, with several minor changes related to multi-PHY coexistence, frequency diversity, and other PHY-MAC interfaces. Referring to Fig. 1, the elements in the SUN usage model can be mapped individually to the components defined in TG4g. Correspondingly, the data collectors, electric/gas/water meters in Fig. 1, can in TG4g be mapped as PANC and normal devices. The mobile data collector may be either a PANC or a device. IEEE 802.11ah — IEEE 802.11ah (TGah) [2] is a TG in the 802.11 wireless local area network (WLAN) WG. Its scope is enhancing MAC and PHY designs to operate in the license-exempt bands below 1 GHz for smart grid and smart utility communications. Most of the frequency bands listed in Table 1 are also covered by TGah. TGah is currently in the process of collecting system design proposals.
TV WHITE SPACE COMMUNICATIONS In line with SUN activities, TVWS is another area receiving overwhelming attention. TVWS is the unused TV channels in the very high frequency (VHF) and ultra high frequency (UHF)
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Fixed device
F
Personal/portable device Mode I (client)
Mode II (independent)
Sensing-only
Geolocation awareness
Required (accuracy ± 50 m)
Not required
Required (accuracy ±50 m)
Not required
Geolocation re-establishment
Not required
Not required
Required (once/min)
Not required
Database access
Required (once/day)
Not required
Required (upon location change)
Not required
Available channels
2–51 (except 3,4,36-38)
Power (EIRP)
21–51 (except 36–38)
4W
Spectrum sensing
100 mW Not required
50 mW Required
Table 2. Summary of FCC governing regulations. regions licensed by TV broadcasters and wireless microphones. Secondary wireless services such as WLAN and WPAN were previously not allowed to use these bands, but recently the Federal Communications Commission (FCC) granted usage of TVWS with several conditions and restrictions [3, 4]. The announcement elated the wireless community, and this section presents these positive updates in regulations and standardization activities.
GOVERNING REGULATIONS The U.S. FCC issued a report and order (R&O) [3] in November 2008, then later in September 2010 [4], outlining the governing regulations for unlicensed usage of TVWS. The regulations for unlicensed TVWS devices, also known as TV band devices (TVBDs), to share TVWS with incumbents such as TV and wireless microphones are summarized in Table 2. Many of the following regulations are closely related to the context of this article. Device Classifications — There are two classes of TVBDs — fixed and personal/portable devices (hereinafter, portable devices). Fixed devices operate at a fixed location with a high-power outdoor antenna. Portable devices operate at lower power and could take the form of a WLAN access point or WPAN module in a handheld device. Portable devices are further divided into Modes I and II. Mode I devices are client portable devices controlled by a fixed device or Mode II portable device. Mode II devices are independent portable devices with the ability to access available channels. Transmit Power — The FCC has imposed strict rules on the allowable transmit power of TVBDs. Fixed devices may transmit up to a 4 W equivalent of effective isotropic radiated power (EIRP), with 1 W output power and a 6 dBi gain antenna. Portable devices are allowed to transmit up to a 100 mW equivalent of EIRP, with no antenna gain, except that when operating in a channel adjacent to a licensed user and within the protected area of that service, the allowable transmit power should be limited to 40 mW. Additionally, sensing-only portable devices are allowed to transmit up to 50 mW.
Geolocation and Database Access Requirements — TVBD geolocation awareness is an essential capability required by the FCC regulations. Fixed devices should be professionally installed with geographic coordinates accurate to ±50m. Mode II devices, however, should incorporate a geolocation capability (e.g., via a Global Positioning System [GPS] module) to determine their locations and should re-establish location at least once every 60 s. In addition to geolocation awareness, fixed and Mode II devices are required to access the TV band database over the Internet to determine the locally available list of TV channels. Fixed devices should access the database at least once a day, whereas Mode II devices should do so if they change location during operation by more than 100 m from where they last accessed the database, or each time they are activated from a power-off condition. Mode I devices should be allowed to operate only upon receiving available TV channels from a fixed or Mode II device. Other Optional Requirements — Spectrum sensing was a mandatory requirement in [3] but was relaxed in [4]. It is currently an optional function with the recommended detection threshold down to –114 dBm. Spectrum sensing may still be a means of determining available TV channels for sensing-only devices.
DEVELOPMENT IN STANDARDIZATION INITIATIVES Several standardization initiatives were formed as a response from the wireless community to the opening of TVWS. IEEE 802.11af — In September 2009, a TVWS Study Group (SG) was formed by the IEEE 802.11 WG to investigate the possibility of a WLAN TVWS standard. The TVWS SG held two meetings in September and November 2009, and produced a document known as the Project Authorization Request (PAR). This was then reviewed and approved by the IEEE Executive Committee (EC) in November 2009. As a result of the approval, the TGaf was officially formed in January 2010, with a scope of defining modifications in PHY and MAC layer designs with ref-
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
135
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
SMART UTILITY NETWORKS IN TV WHITE SPACE Mode I client personal/portable device
Mode I client personal/portable device
Sensing-only device
Mode II independent personal/portable device
Mode I client personal/portable device
Fixed device
Sensing-only device
Figure 2. Deployment scenario of TVBDs. erence to the WLAN legacy standard IEEE 802.11 revision 2007 [5] to enable communications in TVWS. IEEE 802.22 — IEEE 802.22 is a draft standard specifying wireless regional area network (WRAN) communication systems operating in TVWS. The formation of the IEEE 802.22 WG in October 2004 was in response to the Notice of Proposed Rule Making (NPRM) issued by the FCC in May 2004 [6]. Its scope is mainly specifying the enabling policies and procedures for WRAN systems to share the spectrum resources in TVWS by employing cognitive radio techniques. IEEE 802.19.1 — IEEE 802.19 is the Wireless Coexistence Technical Advisory Group (TAG) in IEEE 802. The TAG deals with coexistence of the many unlicensed wireless networks in the IEEE 802 family. In March 2009, the 802.19 TAG received an assignment from the IEEE EC Study Group (ECSG) to develop coexistence scenarios and possible coexistence metrics in TVWS, and an SG within the 802.19 TAG was initiated as a result. In September 2009, the 802.19 SG began work on a new PAR to form a new IEEE 802.19.1 TG focusing on coexistence methods in TVWS.
DEPLOYMENT SCENARIOS Based on the regulations outlined earlier, the deployment scenario of the TVWS network can be as illustrated in Fig. 2, which shows the different classes of TVBDs — fixed and portable devices. Fixed devices are allowed to transmit up to 4 W. They can be connected to a Mode I client device either via direct connection or through a Mode II independent device. Mode II devices are allowed to transmit up to 100 mW. Mode I devices should be controlled by a fixed or Mode II device and are allowed to transmit up to 100 mW. There is also a sensing-only device, which may transmit up to 50 mW.
136
Communications IEEE
Despite the fact that SUN and TVWS are independent and uncorrelated technologies, they can be combined to form a hybrid eco-friendly technology that draws out their individual benefits. SUN operates primarily in unlicensed bands, as shown in Table 1, which indicates that traffic congestion and interference can significantly degrade performance since sharing the spectrum with other wireless systems is unavoidable. Therefore, by deploying in TVWS, more spectrum resources can be made available to the SUN system, thus reducing the degrading-impact of congestion and interference. Additionally, the range of the SUN system can also be extended through TVWS with longer reach and higher penetration capabilities compared to spectrum bands in the gigahertz order. In order to utilize the unused spectrum in TVWS, SUN must comply with governing rules and communication protocols specific to accessing it. To achieve this, SUN and TVWS must display a certain level of homogeneity in terms of deployment scenario and system behavior. The following sections discuss the homogeneity between SUN components and devices in a TVWS communication system, as well as opportunities and challenges in deploying SUN in TVWS.
SUN DEVICES TO TVBDS For SUN to be able to utilize the spectrum in TVWS, the SUN deployment scenario must be capable of matching the TVWS communication deployment scenario. In other words, SUN components must be mapped into the TVWS communication system architecture. This section covers the relationship between SUN components and TVWS devices (i.e., TVBDs). Referring to Figs. 1 and 2, the devices in the SUN usage model may be mapped into different classes of TVBDs, as shown in Table 3. In Fig. 1, the utility providers have high-power base stations at headquarters and control centers for establishing metropolitan area networks. These base stations can be viewed as fixed TVBDs, as shown in Fig. 2. These fixed base stations have outdoor antennas, geolocation awareness capability, and the ability to access the TV band database for incumbent protection contours. The fixed base stations are connected to data collectors, each covering a smaller area, which may be either access points (APs as in IEEE 802.11) or PANCs (as in IEEE 802.15.4 or 802.15.4g). Alternatively, simplified models of data collectors can also be deployed in individual households. The data collectors can be viewed as Mode II independent TVBDs from the perspective of TVWS regulations. The data collectors are connected to the customer’s premises, each with multiple smart meters of different utility services. The smart meters are equivalent to Mode I client TVBDs. In times of link outage, mobile data collectors may be deployed in place of fixed data collectors. A mobile data collector can be viewed as a sensingonly device with very low power and requiring no geolocation awareness capability.
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
F
SUN components
TVBD classes
Sensing capability
Geolocation awareness
Transmit power
Utility provider base station
Fixed device
Not required
Required
4W
Data collector
Mode II independent device
Not required
Required
100 mW
Electric/gas/water meter
Mode I client device
Not required
Not required
100 mW
Mobile data collector
Sensing-only device
Required
Not required
50 mW
Table 3. Mapping from SUN components to TVBD classes.
POTENTIAL ADVANTAGES The previous section shows that the SUN components match well with the device classes in TVWS regulations. This means that the TVBD network may effectively be deployed in the same network topology as the SUN usage model, indicating that the TVWS spectrum can be potentially utilized by the SUN system. Table 3 shows a one-to-one mapping scheme between SUN devices and TVBDs. This section provides some further analysis on the opportunities when the SUN system is implemented occupying the TVWS spectrum. Fundamentally, availability of the TVWS provides additional spectrum resources to SUN systems allocated in the current unlicensed bands. Longer range and higher penetration capabilities are the primary advantages offered by VHF/UHF signals. compared to popular wireless unlicensed bands such as 2.4 and 5 GHz, VHF/UHF bands are significantly more suitable for applications intended for large area coverage with obstacles. For SUN applications, it is essential for the radio sphere of influence to cover a large and population-density-diverse area, from compact metropolitan areas filled with tall buildings to rural areas characterized by sparse distance between homes, hills, and heavy foliage. Existing infrastructure such as telephony and powerline communications has also been found incapable of meeting the SUN’s needs. In this sense, TVWS is a good candidate for extending the SUN operating range. By employing TVWS, a relatively larger area can be covered by a given node, thereby reducing the amount of required infrastructure in a specific area. In other words, installation and maintenance costs can be drastically reduced. From this perspective, TVWS is a suitable candidate for accessing rural areas. TVWS transmission’s higher penetration characteristic also indicates that for indoor connectivity through walls, the number of APs or PANCs can be effectively reduced. From this perspective, TVWS is also advantageous in inner city apartments and offices. Global compatibility of SUN devices is another opportunity offered by taking advantage of TVWS in SUN applications. Currently, different regulatory domains specify respective spectrum bands for these applications. From Table 1, it is clear that most of these allocated bands do not overlap, indicating that the devices must be designed separately to accommodate respective operating bands across the globe. Also, in many realistic scenarios, each regulatory domain will have its own separate market segment, thus
reducing the room for global compatibility. As an example, SUN component manufacturers will have to build separate systems for U.S. devices operating in the 902–928 MHz band and European ones operating in the 863–870 MHz band. By opening TVWS for SUN devices, manufacturers can concentrate on designing devices that commonly operate in TVWS within the VHF/ UHF bands. This will give a more globalized instead of segmented market, thereby improving compatibility among devices from multiple vendors. Apart from that, globalized compatibility of devices also increases competition among vendors and stimulates healthy growth of the SUN industry as a whole.
RISKS: REGULATORY AND TECHNICAL There are also challenges along the path to incorporating TVWS for SUN applications. Among these, the regulatory requirements for TVWS usage are the major concerns for the SUN usage model. As described earlier, Mode II devices are required to have geolocation awareness capability with accuracy of ±50 m. Location must also be re-established every 60 s. This requirement implies that in every Mode II device there must be a geolocation device such as a GPS module. Although geolocation can be effectively determined via GPS in outdoor operations, accuracy is reduced significantly for indoor operations. Mode II devices such as the SUN data collector may be installed outdoors or indoors depending on factors such as physical location and cost. The geolocation awareness capability requirement thus reduces the flexibility of SUN data collector deployment. Ultimately, since the availability of TV channels does not change in a matter of seconds, the requirement may be imposing an overly strict burden on the SUN system as a whole. The issue of license exemption in the SUN frequency bands is an emerging topic. For SUN applications, service reliability is of the utmost importance. This is reasonable since the customer pays for utility services, which makes “best effort connectivity” in conventional secondary wireless services less attractive. In unlicensed bands such as TVWS, reliability is therefore a fundamental concern. Unlicensed operation suggests that a SUN network in TVWS must cease all operations upon discovering the existence of an incoming incumbent service. All these scenarios paint an undesirable picture for utility services. While diversity algorithms such as dynamic frequency switching and multichan-
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
137
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
The main challenge is incorporating two separate technologies into a hybrid system. SUN technology is a low-powerconsumption, low-data-rate, and large-scale network with potentially millions of nodes, whereas TVWS technology is a spectrum accessing technology requiring external network connectivity for authorization.
nel utilization may provide solutions to the problem, the “secondary” characteristic of TVWS is the fundamental issue. Currently, discussions on licensed utility bands are becoming popular in groups such as IEEE 802.15.4g, indicating that utility providers are eager to find solutions that achieve more reliable services. However, it should be noted that this is a typical but not unique problem in TVWS since most frequency bands allocated to SUN currently share the same unlicensed characteristic. From a technical standpoint, the main challenge is incorporating two separate technologies with respective characteristics into a hybrid system. SUN technology is a low-power-consumption, low-data-rate, and large-scale network with potentially millions of nodes, whereas TVWS technology is a spectrum accessing technology requiring external network connectivity for authorization. Each technology has a significant level of behavior and demands not shared by the other. For example, a SUN data collector may need to be deployed as a battery-powered lowduty-cycle indoor device, but a TVWS Mode II device requires a GPS module effective only in an outdoor environment and is mandated to frequently power up to register its location. In this case there is a level of incompatibility between the two that needs to be addressed in the process of combining the two technologies. The former is under development by the TG4g, while the latter is under the watch of the TGaf. Efforts to merge the two technologies are expected to be challenging considering that TG4g and TGaf are going in separate and orthogonal directions.
TOWARD THE FUTURE OF SUN IN TVWS The future of SUN and TVWS technologies is encouraging, and there is little limit to their potential. Yet several crucial elements need to be addressed in order for SUN to be able to fully utilize the advantages of TVWS. From a regulation standpoint, two recommendations are provided. First, requirements specified by regulators for occupying TVWS should be relaxed. Specifically, it is vitally important to relax the geolocation awareness capability for TVBDs intended to support simple and low-cost design (e.g., SUN wireless household data collectors). For SUN applications, this is the bottleneck that dictates AMI cost effectiveness and technical feasibility. Second, the TVWS licensing issue should be considered. Realizing a licensed SUN band may still be premature, but as an alternative, TVWS communications may consider importing the idea of “light licensing,” where a time-location-dependent license is granted to an “enabling TVBD,” which possesses the authority to enable communications of a client or dependent devices without having database access performed locally. The light licensing concept is specified in the 3.6 GHz band long-range WLAN in the United States [7]. On the other hand, from a technical standpoint, two recommendations are provided. First, the differences in respective system demands must be aligned. For example, the conflict
138
Communications IEEE
A
BEMaGS F
between low power consumption in SUN device requirements and a frequent power-up mode for location re-establishment in TVWS Mode II device requirements can be solved by replacing household indoor data collectors with powersource-connected outdoor Mode II data collectors capable of covering an entire neighborhood. Another example is aligning SUN channels for optimized occupancy in TVWS. SUN communication channels are typically several tens to hundreds of kilohertz, while TV channels may span across 6–8 MHz. With careful consideration, multiple SUN channels can be designed to fit into a TV channel to increase spectrum usage efficiency. In standardization activities, efforts on aligning the technologies should take place separately in respective TGs with mutual understanding and collaboration. Second, instead of merging two separate technologies, another recommended effort is a new system design that includes elements from both. This method reduces the amount of modifications to existing specifications in the ongoing TGs. The 802.15 WG recently took the first step in exploring opportunities in applications of smart grids and smart utilities by occupying TVWS. An SG has been formed in the WG to investigate this topic. A new TG is being formed to specify a hybrid system capable of extracting advantages from both SUN and TVWS technologies.
CONCLUSION This article presents the combination of two emerging eco-friendly communication systems — SUN and TVWS. The hybrid system could be capable of conserving energy and simultaneously reusing spectrum resources. The relationship between the characteristics of two systems, current technology trends, and opportunities and challenges in realizing the hybrid system are discussed. As a conclusion, it is observed that while encouraging advantages can be obtained by combining SUN and TVWS technologies, considerable effort is required to make that a reality.
REFERENCES [1] IEEE Std. 802.15.4, “Wireless MAC and PHY Specifications for Low-Rate WPANs,” 8 Sept., 2006. [2] IEEE, “Sub 1 GHz license-exempt PAR and 5C,” https:// ____ mentor.ieee.org/802.11/dcn/10/11-10-0001-13-0wng_____________ 900mhz-par-and-5c.doc [3] FCC, Second Report and Order and Memorandum Opinion and Order: In the Matter of Unlicensed Operation in the TV Broadcast Bands, Doc. 08-260, Nov. 14, 2008. [4] FCC, Second Memorandum Opinion and Order, Doc. 10-174, Sept. 23, 2010. [5] IEEE Standard 802.11 Revision 2007, “Wireless LAN MAC and PHY Specifications,” 12 June 2007. [6] FCC, Notice of Proposed Rule Making: ET Docket no. 04-113, May 25, 2004. [7] IEEE Std. 802.11y, “Wireless LAN MAC and PHY Specifications. Amendment 3: 3650-3700 MHz Operation in USA,” 6 Nov. 2008.
BIOGRAPHIES CHIN-SEAN SUM (
[email protected]) ________ received his M.E. from the University of Technology of Malaysia (UTM) in 2002, and his Ph.D. from Niigata University, Japan, in 2007. In June 200, he joined the National Institute of Information and Communications Technology (NICT), Japan, as an expert researcher in the Ubiquitous Mobile Communications
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Group (UMCG). He is actively involved in the IEEE 802.15.3c (TG3c) standardization activities for millimeter-wave WPAN, where he served as the task group secretary and technical assistant editor for the draft standard. He is the recipient of the IEEE Working Group Chairs Awards for the IEEE 802.15.3c Standard. He is currently the coexistence subeditor in IEEE 802.15.4g WPAN Smart Utility Networks (SUN) and an active contributor in IEEE 802.11af, TV White Space for WLAN. HIROSHI HARADA is director of the Ubiquitous Mobile Communication Group at NICT and is also the director of NICT’s Singapore Wireless Communication Laboratory. He joined the Communications Research Laboratory, Ministry of Posts and Communications, in 1995 (currently NICT). Since 1995, he has researched software defined radio (SDR), cognitive radio, dynamic spectrum access network, and broadband wireless access systems on the microwave and millimeterwave bands. He also has joined many standardization committees and fora in United States as well as in Japan and fulfilled important roles for them, especially IEEE 802.15.3c, IEEE 1900.4, and IEEE1900.6. He serves currently on the board of directors of the SDR Forum and as chair of IEEE SCC41 (IEEE P1900) and vice chair of IEEE P1900.4. He was chair of the IEICE Technical Committee on Software Radio (TCSR), 2005–2007. He is also involved in many other activities related to telecommunications. He is currently a visiting professor at the University of Electro-Communications, Tokyo, Japan, and is the author of Simulation and Software Radio for Mobile Communications (Artech House, 2002). FUMIHIDE KOJIMA [M] received B.E., M.E. and Ph.D. degrees in electrical communications engineering from Osaka Uni-
IEEE
BEMaGS F
versity, Japan, in 1996, 1997, and 1999, respectively. Since he joined the Communications Research Laboratory, Ministry of Posts and Telecommunications in 1999, he has been engaged in research on ITS telecommunications, ROF multimedia transmissions, mobile ad hoc network for disaster radio, and wireless grid systems including smart utility networks. Currently, he is a senior researcher at the new generation wireless research center of NICT. His current research includes intelligent MAC protocol for radio communication systems including mobile ad hoc networks and smart utility networks. ZHOU LAN received his B.S. and Ph.D. degrees in electrical engineering from Beijing University of Posts and Telecommunications (BUPT), China, in 2000 and 2005, respectively. He is currently with NICT as an expert researcher. His research interests include high-speed wireless MAC design, large-scale simulation platform development, and hardware implementation. He servied as TPC vice chair of IEEE PIMRC 2009 and TPC member of IEEE GLOBECOM 2009. He has worked closely with industry. He served as the assistant editor of the IEEE 802.15.3c mmwave WPAN WG. He is also active in other IEEE WLAN and WPAN WGs. R YUHEI F UNADA received his B.E., M.E., and Ph.D. degrees from Chuo University, Tokyo, Japan, in 2000, 2002, and 2005, respectively. From 1999 to 2005 he was a trainee at NICT. In 2005 he joined NICT as a postdoctoral researcher, and is currently a permanent researcher. His research interests include OFDM-based mobile telecommunication systems, single-carrier WPAN systems, and various radio transmission techniques. He received the Young Researcher’s Encouragement Award of IEEE VTS Japan in 2002, and the Best Paper Award of WPMC 2006.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
139
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
ACCEPTED FROM OPEN CALL
Advances in Mode-Stirred Reverberation Chambers for Wireless Communication Performance Evaluation Miguel Á. García-Fernández, Juan D. Sánchez-Heredia, Antonio M. Martínez-González, and David A. Sánchez-Hernández, Universidad Politécnica de Cartagena Juan F. Valenzuela-Valdés, EMITE Ing
ABSTRACT Reverberation chambers (RC) are a popular tool for laboratory wireless communication performance evaluation, and their standardization for Over-The-Air (OTA) measurements is underway. Yet, the inherent limitations of singlecavity RCs to emulate isotropic Rayleigh-fading scenarios with uniform phase distribution and high elevation angular spread put their representation of realistic scenarios into jeopardy. Recent advances in the last few years, however, have solved all these limitations by using more general mode-stirred reverberation chambers (MSC), wherein the number of cavities, their stirring and coupling mechanisms, and their software postprocessing algorithms is far from simple, representing a new era for wireless communications research, development, and over-the-air testing. This article highlights recent advances in the development of second-generation mode-stirred chambers for wireless communications performance evaluation.
INTRODUCTION A reverberation chamber (RC) is a highly conductive enclosed cavity typically equipped with metallic paddles and turntables. The independent movement of paddles and turntables dynamically changes the electromagnetic field boundary conditions. In this way the natural multimode electromagnetic environment inside the single cavity is stirred. With this continuous mode stirring in time, the chamber provides the same statistical distribution of fields independent of location, except for those observation points in close proximity to walls and nearby objects. This required field uniformity also implies polarization balance in the chamber. At any observation point within the chamber, the field will vary from a maximum to a minimum as the different elements (stirrers and turntables) change the boundary conditions [1]. The standard deviation of the mean field throughout the chamber is typically the figure of merit used to assess the per-
140
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
formance of the RC. In a perfectly-stirred RC, the real and imaginary parts of the rectangular components of the electric and magnetic field throughout the chamber are Gaussian distributed, independent with identical variances. Thus, the electric or magnetic field inside a perfectlystirred RC follows a single-cluster Rayleigh probability density function in amplitude and uniform distribution of phase, which resembles the multipath fading in urban scenarios of wireless communications systems. If we assume that the introduction of a matched antenna does not perturb the preexisting field distribution within the chamber, the power received by this matched antenna inside the RC is independent of the antenna gain, directivity, or equivalent area [2]. This, along with the repeatability and reliability of the stochastic reference fields emulated in the RC, makes them ideal candidates to evaluate antenna radiated power for wireless communications systems. Since for handheld wireless communications systems antenna radiated power-related parameters such as Total Radiated Power (TRP) and Total Isotropic Sensitivity (TIS) are the standardized figures of merit, RCs have become a popular tool for evaluating wireless communication performance. Yet, propagating scenarios experienced by users outdoors rarely follow the behavior of a uniform Rayleighfading scenario with single-cluster isotropic scattering. A single-cluster assumes that waves that are reflected or diffracted at the receiver and propagated toward the receiver are grouped into just one collection, corresponding to a group of buildings or objects in a room. In urban environments, for instance, one can find several buildings on both sides of the street and each of them can be modeled as a cluster of scatters. Hence, to describe properly this scattering environment, multiple clusters are needed. An isotropic scattering scenario, also known as uniform, assumes that all angles of arrival at the receiver have equal probability, that is, there is no preferred direction of upcoming waves. A distribution of scatters that leads to a uniform distribution of angles of arrival is also difficult to justify in prac-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
IEEE
tice [3]. In consequence, recent years have witnessed a relatively large number of papers describing novel concepts using more general mode-stirred reverberation chambers (MSCs) [4] with both hardware and software modifications to that of simple single-cavity RCs in order to overcome their innate limitations. In MSCs, the fields do not necessarily have to be constrained to a single cavity or even be provided in a reverberating mode to the researcher. In consequence, MSCs may contain more than one metal cavity that could be coupled through a variety of means, including waveguides, slots or metal plates, among others. Likewise, the shape of these cavities does not have to be restricted to the canonical ones, and additional software control and algorithms allow extraordinary advantages to the researcher over conventional single-cavity RCs. This contribution is a short tutorial that highlights the recent advances in wireless propagation emulation using complex MSCs instead of simple RCs. While only a few of these enhancements have reached commercial stage, the novelty accumulated in the last few years could clearly identify mode-stirred reverberation chambers as a direct competitor of more expensive multiprobe multipath spatial fading emulators using anechoic chambers and an excellent tool for wireless communications R&D processes.
HARDWARE ADVANCES MSCS WITH ENHANCED RAYLEIGH-FADING EMULATION One of the very first enhancements was related to the ability to stir the modes more efficiently. There are many contributions regarding the shape and size of stirrers to ensure quasi-perfect mode stirring. Effective paddles should be large and asymmetrical [3], and some specific shapes have been analyzed [3]. But not only the shapes of the stirrers play a role in the effectiveness of the RC. Beyond the simple linear movements of paddles or circular movement of the turntable typically employed in RCs, recent findings have shown that complex paddle and device-undertest (DUT) movements also provide for some additional enhancements. Both non-linear and complex stirrer movements have been proposed for enhanced field uniformity [5].
MSCS WITH RICIAN-FADING EMULATION One important enhancement is related to the ability to emulate Rician-fading environments. The Rayleigh-fading case (K = 0) typically emulated by an RC is a special case of a more general Rician-fading case (K > 0). The Rician K-factor is defined as the ratio between the power of the coherent component (corresponding to the direct path) over the power of the incoherent component (corresponding to the scattered component) of the received field. In fact, when the RC is not perfectly stirred, the unstirred field component being preserved defines a Rice field in coexistence with the Rayleigh field generated by the stirred components. Stochastic plane wave superposition and separation theories can be employed to obtain
1.0E+02
IEEE
BEMaGS F
0 degrees
1.0E+01
30 degrees
1.0E+00 90 degrees 1.0E-01
1.0E-02
1.0E-03 1000
2000
3000
4000 5000 Frequency (MHz)
6000
7000
Figure 1. Variable K-factor in a mode-stirred reverberation chamber when altering the azimuth orientation of the transmitting antenna [6].
both stirred (equivalent to non-Line of Sight or Rayleigh-fading components) and unstirred contributions (equivalent to Rician-fading components). Yet, in most cases the separation of these two components is aided by employing an excitation source that is pointed toward the DUT, and then it is assumed that all wall reflections interact with the paddles [6]. With only one transmitting antenna, other ways of controlling the K-factor are now possible in an MSC. This includes that the transmitting antenna, with a well-defined radiation patter (azimuth change), can be rotated with respect to the DUT, altering the distance between the transmitting antenna and the DUT (distance change), changing the polarization orientation of the transmitting antenna (polarization change), or varying the cavity’s Q-factor by chamber loading (Q-factor change) [6]. Some variable K-factor results in [6] are illustrated in Fig. 1. If two transmitting antennas are used, a wide range of K-factors can be obtained by pointing one of them toward the DUT and the other one toward the stirrers [6]. Interestingly, the K-factor obtained in a mode-stirred reverberation chamber has also been found to be dependent on the number and position of absorbers placed within the main cavity [7].
MSCS WITH HYPER-RAYLEIGH-FADING EMULATION While Rayleigh and Rician fading are commonly used in wireless propagation emulation, smallscale fading encountered in several new scenarios such as vehicle-to-vehicle systems present frequency-dependent and spatially-dependent fading whose severity exceeds that predicted by the Rayleigh fading model. These scenarios are coined as Hyper-Rayleigh, and a very recent paper has been able to accurately emulate these scenarios using a modified reverberation chamber [8]. In [8], an electrically switched multi-element antenna array was added to an RC, and the enclosure size was made considerably smaller than conventional RC for the same tested fre-
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
K-factor
Communications
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
141
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Reflective blade Stepper motor
TX
RX
Switches
PC Splitter 1:4
Vector network analyzer
Figure 2. Block diagram and picture of modified RC in [8] including electrically controlled fading mechanisms and small size.
-15 -20
Rayleigh
-25
S21
-30 -35 -40 Ricean
-45 -50 -55
Hyper-Rayleigh 2.4
2.41
2.42
2.43
2.44 Frequency
2.45
2.46
2.47
2.48 x 109
Figure 3. Plots of signals that experience Rayleigh, Ricean, and Hyper-Rayleigh fading in [8]. quency range. Figure 3 depicts the plots of signals experiencing Rayleigh, Ricean, and HyperRayleigh fading scenarios in the MSC of [8].
MSCS WITH NON-CANONICAL CONFIGURATIONS By carefully controlling the excitation source of an RC, the homogeneity and isotropic characteristic of the field at a specific position can be controlled. The key to obtain enhanced performance is the ability to shift and weight each mode within the chamber, and an array of exciting antennas was proposed to alleviate the mechanical requirements of RCs [9]. This is straightforwardly derived if one takes into account the fact that the field strength at any observation point within the chamber can be obtained by the integration in the source.
142
Communications IEEE
Changing the sources therefore changes the resulting field strengths. This particularly useful advance has even made researchers coin new terms for MSCs, such as scatter-field chamber or source-stirred chamber, among others. In order to excite additional transversal electromagnetic modes, other non-canonical chamber configurations have been proposed. By exciting the chamber with transmission lines [10], for example, new TEM modes that are transversal to those wires can be excited, further increasing the frequency range of operation. In particular, for the same cavity size, the lowest usable frequency becomes smaller. Different wire and phase shift excitations are also possible. Other noncanonical configurations include those contributions that employ a variable geometry, a moving wall [11], or non-parallel walls [12]. In such non-canonical MSCs no eigenmodes exist and a diffuse, statistically uniform field is created without the use of a mechanical mode stirrer. As a result, test times can be drastically reduced. One recent contribution for enhanced emulation using MSCs is the opening of the door [13]. The aperture of the door transforms one wall that was perfectly electric into a perfectly magnetic wall, but at the same time with a varying aperture degree. Some modes will try to propagate through the opening, and therefore the chamber can no longer be called a reverberation chamber as both reverberating and non-reverberating modes exist. In this way, non-isotropic fading emulation can also be performed using a mode-stirred reverberation chamber, providing for a different number of multipath components (MPC), angle of arrival (AoA), or angular spread values (AS) of the emulated scenarios. Furthermore, the opening of the door can be used for enhancing the accuracy of the chamber for performing antenna radiated power measurements by a more accurate characterization of losses in the chamber by this opening of an aper-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
0
1 absorber: τrms=187 ns
BEMaGS F
100
3 absorber: τrms=106 ns
-5 -10 -15 7 absorber: τrms=66 ns
-20
BER
Power delay profile (dB)
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
10-1
-25 Oil refinery: 3 different positions τrms=103 ns τrms=102 ns τrms=112 ns
-30 -35 -40 0
Real environment Reverberation chamber with no absorber Reverberation chamber with two absorbers Through connection 10-2
0
2
4
50 100 150 200 250 300 350 400 450 500 Time (ns)
6
8 10 12 Signal-to-noise ratio
14
16
18
20
Figure 4. Different PDPs (up) and BER (down) measured using an MSC [14]. ture [2]. It seems that the door of the MSC has been really open, in the wide sense of the word. With the available manipulation of diverse spatial fading multipath characteristics using MSCs, another important step was the ability to control the time-dependent fading performance by being able to emulate variable root-mean square (RMS) delay spreads. Effects such as Doppler spread and fading, which are a consequence of a dynamically moving environment, can also be emulated inside an MSC by moving the paddles with different speeds or using them in stepped or non-linear modes. With the use of absorbers in [14], different RMS delay spread profiles can also be achieved. The ensemble average of the magnitude squared of the impulse response of the MSC is referred to as the power delay profile (PDP) and it is the way to include effects due to time-varying multipath. The shape of the PDP can have adverse effects on the performance of digital communication systems. The RMS delay spread of the PDP is often used to characterize a wireless communication environment because it is directly related to the BitError-Rate (BER) performance of a channel. The BER is an end-to-end performance measurement that quantifies the reliability of the entire radio system from bits in to bits out. Standardized channel models are typically characterized by RMS delay-spreads. As the RMS delay spread in an MSC has been found to be proportional to the chamber Q-factor for a given frequency, this is yet another sign that very accurate standardized channel fading emulation is possible with MSCs. This includes emulating the behavior of the BER for different stirrer velocities [15, 16] and chamber loadings [14, 16], as illustrated in Fig. 4 [14]. While specific power delay profiles can be replicated by adding certain amounts of absorbers and averaging over many different paddle positions, in [17] the transmitter’s excitation signal was injected into a fading emulator prior to introducing it into the chamber. In this way, a channel response having multiple discrete clustered distributions, typically found in both urban and suburban settings where reflecting structures may be located far from the receiver,
Stirrer
Waveguide
Patch antenna
Figure 5. An MSC with two coupled cavities [18]. was created. Very accurate emulation of these realistic environments can be performed using the method described in [17]. A clear advantage of this method compared to the one employed in the next section is the use of only one chamber. The disadvantage is clearly the requirement of a fading emulator.
MSCS WITH MULTIPLE CAVITIES Another important advance is the use of multiple cavities in order to provide for some control of a complex multipath environment consisting on diverse clusters with different fading characteristics. A possibility is to use a metal plate with different-size irises separating two cavities. This can give some control over which modes are coupled to the main cavity and also enlarge the delay spread at the main cavity in comparison to single-cavity RCs [13]. Another possibility is to connect two cavities with waveguides or wires [18], as illustrated in Fig. 5. With this modification, the rank channel can be altered, and complex MIMO fading characteristics such keyholes can also be emulated. This enriches the emulating possibilities of MSC, which now include the ability to emulate degenerated H matrices as it happens in tunnels, for example. With multiple cavities, not only the propagation characteristics
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
143
A
BEMaGS F
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
r2
IEEE
1.0
1.0
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.0
r2
Communications
0.0
-0.2
-0.2
-0.4
-0.4
-0.6
-0.6
-0.8
-0.8
-1.0 -1.0
0.0 r1
-0.5
0.5
1.0
F
-1.0 -1.0
-0.5
0.0 r1
0.5
1.0
Figure 6. Different coherence times at GSM1800 emulated in an MSC [15].
SOFTWARE ADVANCES 30
25
Capacity (bit/Hz/s)
ACCURATE CONTROL OVER MSC ELEMENTS
K=0.001 K=0.01 K=0.098 K=0.33 K=4.26 K=49 K=0.001 [14] K=0.01 [14] K=0.098 [14] K=0.33 [14] K=4.26 [14] K=49 [14]
20
15
10
5
0
0
5
10
15 SNR
20
25
30
Figure 7. Outdoor-measured and MSC-measured MIMO capacity vs. SNR for three different 3 × 3 MIMO systems using the offset technique [24]. of the transmitter and receiver can be modified independently, but MSCs can also reduce the typically high elevation angular spread of RCs. Variable RMS delay spreads have also been obtained with coupled cavities, which have demonstrated their ability to emulate indoor environments, wideband in-vehicle environments [19], or metallic windows, tree canopies, walls and other artifacts in buildings [20]. Interestingly enough, it has been found that for a typical metal-framed window structure, the MIMO capacity is greater than that without metal frames. For an 8 × 8 antenna system, the MIMO capacity is increased by about 2.5 times when metal frames are introduced, and the presence of leaves increases that capacity even more when the transmitted power is kept constant [20]. These enhancements have paved the way for new MSC testbeds for MIMO systems able to emulate standardized fading channels.
144
Communications IEEE
If hardware advances are impressive, progress in software post-processing techniques using MSCs does not fall behind. In [15, 16], for instance, an accurate control of the coherence time of an MSC is achieved by means of a properly tailored modulation of the stirrer velocity, as depicted in Fig. 6. The coherence time is the time over which the channel can be assumed constant. This opens the door for very realistic emulation of the time variability of real propagation channels for wireless device performance and signal propagation testing. The coherence time is the most useful parameter for describing this frequency dispersiveness in the time-domain. Another good and useful advance in the field is to emulate multipath fading using a random time-variable phase for every direction of arrival [21], opening the door for complex standardized channel emulation with time- and phase-dependent parameters. A good example of this is the recent ability to measure the radiation patterns of antennas using a modified reverberation chamber [22]. In [22], the free-space field radiated by the antenna is retrieved from measurements in an MSC and time-reversal techniques. Accuracies typically better than 1 dB over the main lobes were achieved. Since more complex testing tools based on near-field and anechoic chamber methods use the radiation patterns of antennas to estimate the correlation properties and from these properties estimate the MIMO parameters, it seems clear that MSC can soon achieve the same level of performance as more complicated two-stage or multiple test probe methods.
STOCHASTIC HANDLING OF MEASURED DATA SAMPLES A generalized stochastic field model capable of ensuring a continuous transition among very different scattering scenarios by a K-generalized PDF in an MSC has been readily available since 2004 [23]. The application of stochastic sample
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
handling for mode-stirred reverberation chambers has not been suggested until recently [2426]. Stochastic handling of the measured data set of samples is perhaps the most promising technique for further enhancements. For example, the use of an offset technique within the set of measured samples has been reported to emulate Rician-fading very accurately without any hardware change [24]. For a target K-factor, the required offset has to be defined in terms of the radius of cluster data, the distance of centroid of cluster from the origin, and its phase-coherence to the selected radius. A comparison between Rician-fading emulation using this technique and outdoor Rician measurements can be observed from Fig. 7. Good agreement is observed. One very recent method that is able to emulate arbitrary fading scenarios is the sample selection technique [25]. The sample selection technique consists of selecting the sample subset that conforms to a specific target fading statistical ensemble from the whole sample set measured in the MSC using genetic algorithms. It has to be mentioned that the selection is possible because the originally-measured Rayleighfading set is composed of many different clusters due to the multiple-cavity slots-coupled system employed. Only with stochastic theory, this method can really target the emulation of standardized channel models (GSCM, SCM, SCME, Winner-II or IMT-Advanced), which is no longer unheard of for MSCs, as illustrated in Fig. 8. In this figure, the stand-alone normalized 1 × 2 MIMO throughput (spectral efficiency) for a IEEE 802.11n device measured in the E200 MIMO Analyzer by EMITE Ing (illustrated in Fig. 9) following the procedure in [27] is compared to the 1 × 2 802.11n MIMO capacity (Shannon) measured in the E200 by EMITE Ing with the sample selection method. The equivalent spectral efficiency calculated with the public Matlab™ code for the standardized IEEE 802.11n is also depicted for comparison purposes. The 802.11n target data sample for the sample selection algorithm was a 2 × 2 MIMO system at a frequency of 2.4 GHz with nine propagation paths in an office environment (indoor). The E200 MIMO Analyzer is a twocavity MSC with dimensions of 0.82 m × 1.275 m × 1.95 m, eight exciting antennas allowing accurate source-stirring, polarization stirring due to aperture-coupling and to the different orientation of the antenna exciting elements, three mechanical and mode-coupling stirrers, one holder-stirrer and variable iris-coupling between the two cavities. Distribution fitness errors below 2*10 –4 were achieved in less than 40 seconds using a hybrid linear-genetic algorithm. Despite the initial method constraints (the target distribution has to have the same mean power as the initial distribution), it is clear that unheard-of emulation possibilities are provided by the sample selection technique using MSCs. The possibilities for arbitrary emulation and testing in all spatial-, time- and codedomains are very interesting, and MSCs could really equal the performance of more complex spatial-fading emulators based on anechoic chambers at a fraction of the cost in the very near future.
IEEE
BEMaGS F
8 30
Initial distribution Final distribution Target distribution
25
7 6
20 15 10 5 0
5
0
4
0.1 0.05 Amplitude
0.14
3 2 1x2 802.11n Channel model 1x2 802.11n Meas RF device sample selection 1x2 802.11n Meas device 64QAM BW=20MHz 1x2 802.11n Meas device 64QAM BW=40MHz 15 20 SNR (dB)
1 0
5
10
Figure 8. Stand-alone normalized 1 × 2 MIMO throughput for a IEEE 802.11n device measured in the MIMO Analyzer [26].
Switch 2xT
Tx1 Rad com tester
Txn
N Tx DUT
VNA N Rx
SMA Rx1
SMA Rxn
Switch 2xR
Figure 9. The two-cavity MIMO Analyzer MSC [courtesy of EMITE Ing].
CONCLUSIONS In the last few years, different advances have enabled mode-stirred reverberation chambers (MSCs) to solve the inherent limitations of conventional single-cavity reverberation chambers (RC) for wireless communication performance evaluation. It is now clear that MSCs have considerably improved the Clarke’s model followed by conventional single-cavity RCs, and that with arbitrary fading emulation using second-generation MSCs, a new era has started for MIMO research, development, and OTA testing.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
PDF f(x)
IEEE
Spectral efficiency / MIMO capacity (bits/s/Hz)
Communications
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
145
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
REFERENCES [1] Electromagnetic Compatibility (EMC) Part 4-21: Testing and Measurement Techniques Reverberation Chamber, IEC Standard 61000-4-21, 2003. [2] P. Corona et al., “Performance and Analysis of a Reverberating Enclosure with Variable Geometry,” IEEE Trans. Electromagnetic Compatibility, vol. 22, no. 1, 1980, pp. 2–5. [3] J. Clegg et al., “Optimization of Stirrer Designs in a Reverberation Chamber,” IEEE Trans. Electromagnetic Compatibility, vol. 47, Nov. 2005, pp. 824–32. [4] T. A. Loughry and S. H. Gurbaxani, “The Effects of Intrinsic Test Fixture Isolation on Material Shielding Effectiveness Measurements Using Nested Mode-Stirred Chambers,” IEEE Trans. Electromagnetic Compatibility, vol. 37, no. 3, 1995, pp. 449–52. [5] P. Plaza-Gonzalez et al., “New Approach for the Prediction of the Electric Field Distribution in Multimode Microwave-Heating Applicators with Mode Stirrers,” IEEE Trans. Magnetics, vol. 40, no. 3, May 2004, pp. 1672–78. [6] C. L. Holloway et al., “On the Use of Reverberation Chambers to Simulate a Rician Radio Environment for the Testing of Wireless Devices,” IEEE Trans. Antennas and Propagation, vol. 54, no. 11, 2006, pp. 3167–77. [7] A. Sorrentino et al., “The Reverberating Chamber as a Line-of-Sight Wireless Channel Emulator,” IEEE Trans. Antennas and Propagation, vol. 56, no. 6, June 2008, pp. 1825–30. [8] J. Frolik et al., “A Compact Reverberation Chamber for Hyper-Rayleigh Channel Emulation,” IEEE Trans. Antennas and Propagation, vol. 57, no. 12, Dec. 2009. [9] J. S. Hong, “Multimode Chamber Excited by an Array of Antennas,” Electronics Letters, vol. 22, no. 19, 1993, pp. 1679–80. [10] D. Weinzierl et al., “Numerical Evaluation of Noncanonical Reverberation Chamber Configurations,” IEEE Trans. Magnetics, vol. 44, no. 6, June 2008, pp. 1458–61. [11] Y. Huang and D. L. Edwards, “An Investigation of Electromagnetic Fields Inside A Moving Wall Mode-Stirred Chamber,” Proc. 8th IET Int’l. Conf. Electromagnetic Compatibility, 1992, pp. 115–19. [12] F. B. J. Leferink, “High Field Strength in A Large Volume: The Intrinsic Reverberation Chamber,” Proc. IEEE Int’l. Symp. Electromagnetic Compatibility, 1998, pp. 24–27. [13] J. F. Valenzuela-Valdés et al., “Diversity Gain and MIMO Capacity for Non-Isotropic Environments Using A Reverberation Chamber,” IEEE Antennas and Wireless Propagation Letters, vol. 8, 2009, pp. 112–15. [14] E. Genender et al., “Simulating the Multipath Channel with A Reverberation Chamber: Application to Bit Error Rate measurements,” IEEE Trans. Electromagnetic Compatibility, 2010. [15] A. Sorrentino et al., “Characterization of NLOS Wireless Propagation Channels with A Proper Coherence Time Value in A Continuous Mode Stirred Reverberating Chamber,” Proc. 2nd European Wireless Technology Conf., 2009, pp. 168–71. [16] A. Sorrentino et al., “On the Coherence Time Control of A Continuous Mode Stirred Reverberating Chamber,” IEEE Trans. Antennas and Propagation, vol. 57, no. 10, Oct. 2009, pp. 3372–74. [17] H. Fielitz et al., “Reverberation-Chamber Test Environment for Outdoor Urban Wireless Propagation Studies,” IEEE Antennas and Wireless Propagation Letters, vol. 9, 2010, pp. 52–56. [18] M. Lienard and P. Degauque, “Simulation of Dual Array Multipath Channels Using Mode-Stirred Reverberation Chambers,” Electronics Letters, vol. 40, no. 10, 2004, pp. 578–80. [19] O. Delangre et al., “Modeling in-Vehicle Wideband Wireless Channels Using Reverberation Chamber Theory,” Proc. IEEE Vehic. Tech. Conf., Sept. 2007, pp. 2149–53. [20] Z. Yun and M. F. Iskander, “MIMO Capacity for Realistic Wireless Communications Environments,” Proc. IEEE Antennas and Propagation Society Int’l. Symp., June 2004, pp. 1231–34. [21] A. Khaleghi et al., “Evaluation of Diversity Antenna Characteristics in Narrow Band Fading Channel Using Random Phase Generation Process,” Proc. IEEE Vehic. Tech. Conf., 2005, pp. 257–61. [22] A. Cozza and A.e.A. el-Aileh, “Accurate Radiation-Pattern Measurements in A Time-Reversal Electromagnetic Chamber,” IEEE Antennas and Propagation Mag., vol. 52, no. 2, Apr. 2010, pp. 186–93.
146
Communications IEEE
A
BEMaGS F
[23] P. Corona et al., “Generalized Stochastic Field Model for Reverberating Chambers,” IEEE Trans. Electromagnetic Compatibility, vol. 46, no. 4, Nov. 2004, pp. 655–60. [24] J. F. Valenzuela-Valdés and D. A. Sánchez-Hernández, “Emulation of MIMO Rician-Fading Environments with Mode-Stirred Chambers,” accepted for publication at IEEE Trans. Antennas and Propagation, 2010. [25] J. F. Valenzuela-Valdés et al., “Sample Selection Method for Rician-Fading Emulation using Mode-Stirred Chambers,” IEEE Antennas and Propagation Wireless Letters, vol. 9, 2010, pp. 409–12. [26] CTIA Certification Program Working Group Contribution Number RCSG100302. Standardized Fading Channel Emulation for MIMO OTA Using A Mode-Stirred Chamber With Sample Selection Method, Mar. 2010. [27] N. Olano et al., “WLAN MIMO Throughput Test in Reverberation Chamber,” Proc. IEEE Int’l. Symp. Antennas and Propagation, July 2008, pp. 1–4.
BIOGRAPHIES MIGUEL Á. GARCIA-FERNANDEZ was born in Cartagena, Spain. He received the Dipl.-Ing. degree in telecommunications engineering from the Universidad Politécnica de Cartagena, Murcia, Spain, in 2005 and the Ph.D. degree from the Universidad Politécnica de Cartagena, Murcia, Spain, in January 2010. From 2005 onwards, he joined the Department of Information Technologies and Communications, Universidad Politécnica de Cartagena, Murcia, Spain. From October 2009 to September 2010 he also joined the Department of Applied Mathematics and Statistics, Universidad Politécnica de Cartagena, Murcia, Spain. His current research areas cover multiple-input-multiple-output communications, SAR measurements and thermoregulatory processes due to electromagnetic field exposure. J UAN D. S ANCHEZ -H EREDIA was born in Lorca, Spain. He obtained his Telecommunication Engineering Degree from the Universidad Politécnica de Cartagena in 2009 which culminated with the Final Degree Award. In 2007 he worked at General Electric (Cartagena), and was involved in several projects in relation with the network infrastructure. In 2009 he joined the Department of Information Technologies and Communications, Universidad Politécnica de Cartagena (Spain), as a Ph.D. student. Actually he is obtaining a Master Degree in Information Technologies at Universidad de Murcia. His current research areas cover MIMO communications, multimode-stirred chambers and electromagnetic dosimetry. A NTONIO M. M ARTINEZ -G ONZALEZ obtained his Dipl.-Ing. in Telecommunications Engineering from Universidad Politécnica de Valencia, Spain, in 1998 and his Ph.D. from Universidad Politécnica de Cartagena, in early 2004. From 1998 till September 1999, he was employed as technical engineer at the Electromagnetic Compatibility Laboratory of Universidad Politécnica de Valencia, where he developed assessment activities and compliance certifications with European directives related with immunity and emissions to electromagnetic radiation from diverse electrical, electronic and telecommunication equipment. From September 1999 he is an Associate Professor at Universidad Politécnica de Cartagena. Research works developed by him were awarded with the Spanish National Prize from Foundation Airtel and Colegio Oficial de Ingenieros de Telecomunicación de España to the best final project on Mobile Communications in 1999. At present, his research interest is focused on electromagnetic dosimetry, radioelectric emissions and mode stirred chambers. In December 2006 he is one of the founders of EMITE Ing, a technological spin-off company founded by Telecommunication Engineers and Doctors of the Microwave, Radiocommunications and Electromagnetism Research Group (GIMRE) of the Technical University of Cartagena (Spain). Founding of EMITE took place right after the second i-patentes prize to innovation and technology transfer in the Region of Murcia (Spain) was awarded to the company founders. In 2008 GIMRE group was awarded this prize again. J UAN F. V ALENZUELA -V ALDÉS (juan.valenzuela@emite-inge________________ nieria.es) _____ was born in Marbella, Spain. He received the Degree in Telecommunications Engineering from the Universidad de Malaga, Spain, in 2003 and his Ph.D. from Universidad Politécnica de Cartagena, in May 2008. In 2004 he worked at CETECOM (Malaga). In 2004, he joined the Department of Information Technologies and Communica-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
tions, Universidad Politécnica de Cartagena, Spain. In 2007 he joined EMITE Ing as CTO. His current research areas cover MIMO communications, multimode-stirred chambers and SAR measurements. A. S A N C H E Z -H E R N A N D E Z [M’00, SM’06] DAVID (
[email protected]) ____________ obtained his Dipl.-Ing. in Telecommunications Engineering from Universidad Politécnica de Valencia, Spain, in 1992 and his Ph.D. from King’s College, University of London, in early 1996. From 1992 to 1994 he was employed as a Research Associate for The British Council-CAM at King’s College London where he worked on active and dual-band microstrip patch antennas. In 1994 he was appointed EU Research Fellow at King’s College London, working on several joint projects at 18, 38 and 60 GHz related to printed and integrated antennas on GaAs, microstrip antenna arrays, sectorization and diversity. In 1997 he returned to Universidad Politécnica de Valencia, Spain, to the Antennas, Microwaves and Radar Research Group and the Microwave Heating Group. In early 1999 he received the Readership from Universidad Politécnica de Cartagena, and was appointed Vice Dean of the School for Telecommunications Engineering and leader
IEEE
BEMaGS F
of the Microwave, Radiocommunications and Electromagnetism Engineering Research Group. In late 1999 he was appointed Vice Chancellor for Innovation & Technology Transfer at Universidad Politécnica de Cartagena and member of several Foundations and Societies for promotion of R&D in the Autonomous Region of Murcia, in Spain. In May 2001 he was appointed official advisor in technology transfer and member of The Industrial Advisory Council of the Autonomous Government of the Region of Murcia, in Spain, and in May 2003 he was appointed Head of Department. He is also a Chartered Engineer (CEng), IET Fellow, IEEE Senior Member, CENELEC TC106X member, and is the recipient of the R&D J. Langham Thompson Premium, awarded by the Institution of Electrical Engineers (now formerly the Institution of Engineering and Technology), as well as other national and international awards. He has published 3 international books, over 45 scientific papers and over 90 conference contributions, and is a reviewer of several international journals. He holds five patents. His current research interests encompass all aspects of the design and application of printed multi-band antennas for mobile communications, electromagnetic dosimetry issues and MIMO techniques for wireless communications.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
147
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
ACCEPTED FROM OPEN CALL
System-Level Simulation Methodology and Platform for Mobile Cellular Systems Li Chen, Wenwen Chen, Bin Wang, Xin Zhang, Hongyang Chen, and Dacheng Yang
ABSTRACT System-level simulation has been widely used to evaluate the comprehensive performance of different mobile cellular systems. System-level simulation methodologies for different systems have been discussed by different organizations and institutions. However, the framework for a unified simulation methodology and platform has not been established. In this article, we propose a general unified simulation methodology for different cellular systems. Both the design of the simulation structure and the establishment of the simulation platform are studied. Meanwhile, the unified modeling and the realization of various modules related to the system-level simulation are presented. The proposed unified simulation methodology and the general simulation platform can be used to evaluate the performance of multiple mobile communication systems fairly. Finally, the overall performance of LTE and Mobile WiMAX systems is evaluated through the proposed framework. The key simulation results for both Full Buffer and VoIP traffics are presented and discussed. It is shown that the LTE system exhibits better performance than Mobile WiMAX.
INTRODUCTION Mobile communication has continued to evolve rapidly in recent years. The third-generation (3G) mobile cellular systems, such as wideband code division multiple access (WCDMA), cdma2000, time division synchronous code division multiple access (TD-SCDMA), and world interoperability for microwave access (WiMAX), have been commercialized, while the research for the new beyond 3G (B3G) systems (e.g., the Third Generation Partnership Project (3GPP) long term evolution (LTE) and 802.16m) or even 4G systems (e.g., LTE-Advanced (LTE-A)) are still in progress. Many commercial mobile systems today, which are based on orthogonal frequency division multiple access (OFDMA) technology with relatively wide bands, evolved from International Mobile Telecommunications-2000 (IMT-2000). In order to evaluate the expected performance of these mobile cellular systems or to do research on related key technologies of air interface, we
148
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
often have to resort to the system-level simulation. Due to the importance of system-level simulation, multiple simulation methodologies have been proposed by 3GPP, 3GPP2, and the Institute of Electrical and Electronics Engineers (IEEE). Each organization has published the corresponding simulation methodology for the system it standardizes. For example, 3GPP2 provided the simulation methodology for cdma2000 1x evolution-data optimized (EV-DO)/evolution-data and voice (EV-DV) evaluations [1], IEEE announced [2, 3] for 802.16 series standards, and 3GPP issued [4, 5] for WCDMA. Besides, numerous related papers have been published in IEEE. These methodologies evolve with the standardization progress. Each of them, however, focuses only on one specific system. One may ask a natural question: can we evaluate the performance of multiple mobile cellular systems through system-level simulation in a unified framework? Actually, this problem has not been studied extensively. In [6], Gao et al. has proposed a fair manner to evaluate different systems. However, they only unified the simulation configurations to compare the performance of different systems fairly. There are still some issues that need to be clarified regarding the design and realization of a system-level simulation platform and the unified simulation methodology. Also, several other problems about the system-level simulation e.g., how to model the modules and interfaces, how to evaluate the key technologies involved, and the network performance, etc., have not been fully considered yet. All of these obstacles constitute the motivations for our work in this article. Our major task in this article is to establish a unified framework for system-level simulation methodologies for different mobile cellular systems. In addition to capturing important aspects of the unified simulation methodology, this article also highlights the modules in the framework for system-level simulation. Moreover, the issues among different simulation methodologies developed by various standardization bodies are clarified. Finally, we show an example evaluating the performance of LTE and Mobile WiMAX systems using the unified system-level simulation methodology. The analytical and simulation
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
work provides much insight into the technical principle and also the benefits of a deep understanding of the potential of commercial mobile communication systems. The rest of this article is organized as follows. The next section gives an overview of systemlevel simulation. We study the unified modeling of various simulation modules. After that, the unified system-level simulation evaluation methodology and platform are proposed. We present simulation results by using the proposed methodology and platform. Finally, conclusions are summarized.
A
BEMaGS F
Resource management Application layer System-level simulation
MAC layer parameter Traffic model
TCP/IP layer
Transport layer
Interface
MAC layer
OVERVIEW OF SYSTEM-LEVEL SIMULATION PURPOSE OF SIMULATION Due to the complicated structures of mobile cellular communication systems, we cannot describe them completely through a simple and abstracted mathematical model. Thus, we always resort to the simulation to evaluate their performance. Computer programs are used to simulate the operating mechanisms of mobile cellular communication systems, the loaded traffics, etc. The performance of these systems can be reflected by the results obtained from the simulation programs ultimately. Moreover, the simulation can be used as an assistant tool for theoretical studies. Whenever new algorithms or strategies are proposed, we usually cannot apply them directly to real networks to evaluate their performance because of the high cost. Since the simulation is able to emulate the practical scenarios in a statistical manner, qualitative or quantitative complexity analysis and performance evaluation of any new algorithm or strategy can rely on the simulation. In the following, we discuss the advantages of the simulation, including its efficiency and flexibility, which are also the reasons why we need the simulation. The efficiency means that we can develop a simulation platform for a mobile communication system in a very short period of time instead of constructing a practical complicated system. By the use of the simulation platform, the system performance can be predicted easily and effectively. The flexibility refers to facilely changing any component of the system by modifying the corresponding programming module with low cost and risk. Valid conclusions and suggestions can be obtained from the simulation. However, they are not obsoletely accurate since it is impossible for the simulation to describe the physical nature of the actual system with infinite precision. The simulations do not have complete authenticity, reliability, or reproducibility, but they have quasi-authenticity, quasi-reliability, and quasireproducibility. They only approach the physical nature in the sense of probability.
LINK-LEVEL SIMULATION VS. SYSTEM-LEVEL SIMULATION The simulation for mobile communication systems includes the link-level simulation and the system-level simulation. Both of them are widely
Physical layer parameter Physical layer
Link-level simulation
Propagation model Channel model
Figure 1. Component layers and model for simulation methodology. employed to evaluate the associated performance. The link-level simulation focuses on the performance of a transmission between base stations (BSs) and mobile stations (MSs). The performance metrics usually include the bit error rate (BER), signal to noise ratio (SNR), achievable rate, etc. In general, the link-level simulation concentrates on the physical layer. On the left side of Fig.1, we show its relationship to other components in communications. For the purpose of theoretical studies, the performance of modulation/demodulation or coding/decoding schemes in different radio channel models can be obtained from the link-level simulation. The scenario for the system-level simulation generally consists of a network with multiple BSs and MSs. Different from the link-level simulation, the system-level simulation focuses on the application layer performance metrics as expressed by system throughput, user fairness, user-perceived quality of service (QoS), handover delay or success rate, etc. The system-level simulation concentrates on the higher layers above the physical layer, such as the MAC layer, transport layer, TCP/IP layer, and application layer. Figure 1 shows the component layers related to the system-level simulation. For the purpose of theoretical studies, the performance of resource allocation, handover, cell deployment, or other strategies can be obtained from the system-level simulation.
STRUCTURE OF SYSTEM-LEVEL SIMULATION The simulation methodology model is plotted in Fig. 1. System-level simulation includes the scheduling process, power control process, adaptive modulation and coding scheme (MCS) selection process, and other MAC layer processes. Also, the system-level simulation needs to be operated with incorporation of the link-level simulation. In general, the link-level simulation is separately abstracted to a set of SNR-BER curves on different MCS levels. The outputs of the link-level simulation are mapped through an interface to the system-level simulation as inputs of the system-level simulation.
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
149
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Simulation methodologies have been comprehensively discussed in different organizations and institutions. With the evolution of communication standardization, international organizations and companies have made much effort to develop the simulation methodologies for them and their evolutions.
How can we know the precision of this system-level simulation methodology? How can we know whether the proposed platform for the system-level simulation is comparable to other works? In order to guarantee the precision and comparability, we developed the following mechanism. First, we establish the platform for the system-level simulation of a certain cellular system, and then the platform needs to be calibrated with the corresponding standard organizations, such as the simulation methodologies from the International Telecommunication Union Recommendations (ITU-R), 3GPP, 3GPP2, and IEEE WiMAX forum. The calibration method is to adapt the simulation platform to the achievement of the same simulation results as that from the standard organizations with the same parameters and conditions. After the common calibration, we can confirm that the simulation platform is accurate enough to implement the performance evaluation or compare with others.
OVERVIEW OF EXISTING SIMULATION METHODOLOGIES Simulation methodologies have been comprehensively discussed in different organizations and institutions. With the evolution of communication standardization, international organizations and companies have made much effort to develop the simulation methodologies for them and their evolutions. The simulation methodologies for cdma2000 and Mobile WiMAX systems have been explicitly presented in 3GPP2 C.R1002 [1] and WiMAX forum document [3], respectively. 3GPP has also made an effort on simulation methodology for WCDMA and its evolutions such as HSDPA/HSUPA and LTE. However, the methodology for LTE-A has not been established yet. It would be improved with the development of the LTE-A standard. These simulation methodologies involve many modules, including the cell layout model, channel model, radio resource allocation, interference model, physical layer abstraction, traffic models, and other key factors. In the next section, we will study the unified modeling of these modules in detail and discuss their realization or deployment in the system-level simulation.
UNIFIED MODELING OF SIMULATION MODULES Through the overview of existing methodologies, we find that different simulation methodologies and configurations proposed by various organizations aim at different systems, and they are not applicable to each other. Thus, we propose a unified structure for a simulation model and simulation platform, and establish a general framework for different system-level simulation methodologies in this article. The models proposed here provide a unified method to evaluate the performance of various systems. It eliminates many inconsistent aspects in previous literature and keeps the most essential parts. There are many static and dynamic modules for the system-level simulation. In this section,
150
Communications IEEE
A
BEMaGS F
we study the unified models of various modules, which are the components of the unified systemlevel simulation methodology. Their simulation methods and realization in the platform are also discussed.
CELL LAYOUT MODEL AND WRAPAROUND TECHNOLOGY There are several common service area models: single-cell, 7-cell, 19-cell, and 36-cell. If the number of cells in the model is too large, which means the system has many BSs and MSs, it will take a long time to develop and debug the simulation platform. The reason is that the calibration of the simulation platform needs iteratively modifying, running the platform, and analyzing results. Thus, the model with fewer cells can accelerate the evolution of the simulation platform. However, if the number of cells in the model is too small, the inter-cell interference (ICI) suffered by one cell may not be enough. Thus, it may not be able to reflect the practice with tolerable precision. Therefore, the scalability and the adaptability of the simulation platform need to be considered in the selection of the service area model. In some methodologies, the 19cell model is the preferred standard service area. WrapAround is a technology that can simulate the ICI and at the same time improve simulation efficiency. In the 19-cell model without WrapAround, a serious defect would happen in the ICI calculation for the MSs in the edge cells. After the simulation, only the data of the center cell that are reliable can be collected, which leads to low efficiency. Much more time is consumed by the work of the MSs that are out of the center cell, but they only serve as foils to a small quantity of MSs in the center cell. Thus, WrapAround is recommended as the cell layout of the simulation platform to form a toroidal surface to enable faster simulation run times. A toroidal surface is chosen because it can be easily formed from a rhombus by joining the opposing edges. The cell cluster of 19-cell is virtually repeated 8 times at rhombus lattice vertices. Then the structure includes the original 19-cell cluster remaining in the center, called the center cell cluster, while the eight copies evenly surround this center set. By adopting WrapAround technology, the simulation platform can solve the ICI calculation problem since each cell of the center cluster has enough interference. Also, it is able to improve the efficiency of the simulation and the data statistics since all the data in the center cell cluster can be collected. It only makes the system realization a little more complicated. Typically, the MSs are uniformly distributed in the system. Some methodologies [3, 7] specify that every cell or sector has the same number of MSs. This is because with the same number of MSs, not only the traffic density is guaranteed to be uniform in the simulation, but also the effects of the admission control could not be taken into account specially. In order to satisfy the requirements of debugging, testing and researching, the simulation platform should support some fashions of MS dropping. These may include fixed point dropping with only one MS in the service
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
area or fixed point dropping with one MS each sector, which can ensure the fast debugging of the platform; fixed point dropping with multiple MSs in each sector, which is beneficial to the fast comparison testing; hotspot dropping, which can support the research for the hotspot, admission control and congestion control; or dropping with fixed cell radius. It proposes the requirements for the design of the simulation system that the operations of users and the realization of dropping algorithm should be detached.
CHANNEL MODEL The multipath channel model proposed in ITU-R M.1225 [7] has been widely used for the cellular systems with a single-input multiple-output (SIMO) (typically 1 transmit by 2 receive antennas) antenna configuration. M.1225 defines several scenarios: indoor A/B, pedestrian A/B, and vehicular A/B, each with six subpathes except for pedestrian A, which has four subpathes. The relative delay profile and average power for each subpath are also specified. Different systems have distinct bandwidths, which leads to different multipath resolutions, and we cannot use these models directly. Thus, channel models need to be modified for different systems. If multiple-input multiple-output (MIMO) technology is adopted, the spatial-time channel model (SCM) or SCM-Enhancement (SCM-E) [8] should be used to generate multipath fading. Indeed, SCM-E can also be used in SIMO with suitable setting. The simulation platforms for different systems adopt different channel models, which are recommended in corresponding simulation methodologies. For the purpose of the performance comparison among several systems, the unified mixed channel model should be applied. In [6], the authors describe how to unify the model parameters for ITU channels. They established the unified channel model by finding the similarity of number of paths, the power and the delay profile of each path, among different channel models.
RESOURCE ALLOCATION STRATEGY Resource allocation, including subcarrier assignment among users and power allocation on subcarriers, is an important issue in OFDM-based mobile cellular systems. In general, the resource allocation process can be divided into several steps: determination of the available resource; calculation of the packet transmission order over the air interface; implementation of the resource allocation signaling process; and generation of downlink (DL) or uplink (UL) MAC packets. Typical optimization problems of interest for resource allocation include the capacity problem, i.e., maximizing the sum data rate subjects to a power constraint, or the power control problem, i.e., minimizing the transmit power so that a certain quality of service metric for each user is satisfied. Besides, in order to balance the system efficiency and user fairness, we can also try to maximize the aggregate utility of users in the network. More strategies on this issue for OFDM systems are studied in [9] and its references. A basic resource unit is the minimum unit allocated to one user, and is unique for each sys-
tem. In our simulation platform, a basic resource unit in the time-frequency domain is one subchannel in Mobile WiMAX or one resource block (RB) in LTE. On each resource unit, the proportional fair (PF), round robin scheduling algorithm, or other scheduling strategies are used for subcarrier assignment.
IEEE
BEMaGS
INTERFERENCE MODEL Interference is a primary factor that significantly affects system performance in multiuser networks, especially in OFDM systems. One of the key benefits of the OFDMA air interface is its ability to enable frequency reuse, that is, the same frequency can be used in all neighboring cells and sectors. It makes the system deployment much easier since frequency planning is no longer needed. With high frequency reuse patterns, however, the system becomes interference limited. The ICI seen by an MS in DL or a BS in UL is typically frequency- and time-selective. In the system-level simulation, the ICI should be modeled according to the practical channel model, including large scale fading and fast fading components. In our simulation platform, the interference is computed in real-time along with the simulation running through the signals received by the MS from different BSs or by the BS from different MSs.
F
Interference is a primary factor that significantly affects system performance in multiuser networks, especially in OFDM systems. One of the key benefits of the OFDMA air interface is its ability to enable frequency reuse, that is, the same frequency can be used in all neighboring cells and sectors.
PHYSICAL LAYER ABSTRACTION In general, in order to reduce the complexity of the simulation platform, the effects of the linklevel strategies are abstracted to a set of curves, which are the inputs of the system-level simulation as described previously. In OFDM systems, the total bandwidth is divided into a number of orthogonal subcarriers, each of which has a signal to interference plus noise ratio (SINR). Several subcarriers are combined into a subchannel, and its effective SINR is the combination of the SINRs of these subcarriers. Many mapping solutions have been proposed. The 3GPP and the WiMAX Forum AWG group recommend the exponential effective SINR mapping (EESM) model as a typical and default solution for the SINR combination. When the bandwidth is large, such as 10MHz in Mobile WiMAX or LTE systems, the size of fast Fourier transform (FFT) is 1024. The amount of calculation for the SINRs on subcarriers or RBs becomes quite large, which would greatly reduce simulation efficiency. Thus, we propose an interpolation method, which calculates the SINR every four or eight subcarriers first, and then the SINRs on all subcarriers can be obtained through linear interpolation. Through our simulation verification and validation, it can be found that if the effective SINR is calculated by using the values on every four subcarriers instead of every subcarrier, the simulation accuracy under the conditions and models mentioned above will not be affected, but the simulation complexity will be greatly reduced.
SERVICE AND TRAFFIC MODELS With the development of mobile cellular systems, the traffics evolve from voice dominant toward mixed ones. Different types of traffics have dif-
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
151
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Driving module
Forward
iven
)
Slotwork()
Readtime()
k(
Slo tw or k
or
Driven object
tw
Driven object
Slo
()
-dr
e Tim
Driven object
Figure 2. Structure of modules. ferent performance evaluation output metrics. Although there are various types of traffics in practical systems, they can be classified into two categories. One is the packet data traffic without delay requirement, such as Full Buffer. System throughput and user fairness are two guidelines for this type of traffic. However, the throughput is associated with the system bandwidth and typically increases when more bandwidths are occupied. Thus, the spectrum efficiency, which is defined as the throughput divided by the effective bandwidth with unit of bits per second per Hertz, is the normalized performance index for this type of traffic. The other category is the delay sensitive traffic that has a delay requirement, such as voice over IP (VoIP) or streaming media. The packet delay and outage probability are the most critical performance metrics. Different traffic models generate data with different characteristics. However, all traffics can be abstracted as a data generator in the system-level simulation platform. In the platform, there is a data pool at the transmitter which contains data to be transmitted. The data generators for various traffic models fill in this pool with different rules. In practice, the mixed traffic is preferred.
OTHER KEY ALGORITHMS Other key technologies such as hybrid automatic repeat request (HARQ), adaptive modulation and coding (AMC), channel quality indication (CQI) feedback, power control, and interference over thermal (IoT, or rise over thermal, RoT) control in UL are standardized specifically in different systems or widely used. They may be adjusted for different systems according to their own characteristics.
UNIFIED SYSTEM-LEVEL SIMULATION METHODOLOGY AND PLATFORM Based on the unified modeling of various modules, we study the unified system-level simulation methodology and the general simulation plat-
152
Communications IEEE
A
BEMaGS F
form in the following. In order to compare the performance of different systems, the simulation settings should be unified. Since different systems operate under various parameters or configurations, each organization for one system proposes a body of simulation methodology and recommends a set of parameters. They cannot match each other completely. The unified simulation parameters and configurations can be obtained by comparing the characteristics of different systems, as discussed in [6]. For most of the parameters and configurations, we choose a set of common values from the values’ range or optional configurations of different systems. If there is no common value for some parameters or some system specific configurations, we could set them as close as possible for different systems. The system-level simulation platform developed by us is established based on the model shown in Fig. 1. By realizing all the unified modules above, we calibrate the simulation platform by comparing it with published results. After that, the platform can be used for the research or system performance comparison under the unified settings. The unified simulation methodology consists of the unified models of various modules and this unified flow for different systems. The general system-level simulation platform adopts a time-driven mode to drive the system progress, which is the combination of drop and time slot driven. A drop is a process by which all MSs are dropped in the service area in a certain manner. In our platform, the MSs are uniformly distributed in the system. A time slot is the minimum step duration for time in the simulation, which is determined by the system-level minimum control interval. In the Mobile WiMAX system, the period for the scheduling or the power control is a frame, so the time slot in the platform is 5ms, while in the LTE system, the time slot is 1ms since the control period is a subframe. The simulation duration for a drop contains many time slots, which is long enough for the system to become stable. The data for the system outcomes should be collected from the time slot when the simulated system is stable till the end of a drop. A drop can be regarded as one Monte-Carlo simulation. Thus, multiple drops should be simulated in the system-level simulation platform before average values are obtained. The flow for the system-level simulation platform is as follows. Before the time slot simulation, the initialization should be implemented first, which includes reading simulation parameters, initializing service area and BSs, dropping MSs in the system, initializing MSs and the path losses of links between MSs and BSs, etc. After that, the time-driven module drives the progress of the system, and other modules perform their own work in every time slot, which can be summarized in function slotwork(). The progress ends when the simulation time is out. Through the simulation flow, we can observe the relationship among different modules in the simulation platform, which is shown in Fig. 2. The interface between the driving modules and
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
IEEE
the driven objects is the slotwork() function. The clock in Fig. 2 is the time slot module. Only the system driving module that is the central controller can push the clock forward and reset it, while the other modules are driven objects, such as the behaviors of MSs and BSs, which can only read the time slot module. In the whole process of the system-level simulation, all the modules are operating based on the time slot module. The proposed unified system-level simulation methodology and the general simulation platform can be used to evaluate the performance of any mobile communication system or mobile broadband technology. The output performance of different systems is compared fairly and credibly through the unified simulation. In addition, the simulation platform can be used to evaluate the comprehensive performance of the proposed algorithm reflected in the system. With the development of mobile communication standards, the process and method to establish the unified simulation methodology are similar. The unified modeling and methodology can also be used in other areas, such as network planning and network optimization, for which suggestions can be provided through the systemlevel simulation. Moreover, the unified methodology can be extended to compare different network costs by adding some cost modules in the platform. It would be possible for operators to lower their capital and operating expenditures by deploying the system with better performance.
SIMULATION RESULTS In this section, by using the unified simulation methodologies and the methods for the unified parameters studied above, we evaluate the performance of LTE and Mobile WiMAX systems. In the following, the simulation configurations are presented first, and then the results are shown and analyzed. Some evaluating indicators for different traffic models are also presented in the following analysis. By comparing the system parameters and configurations in [3, 10], we summarize a set of unified configurations, as shown in Table 1. The V-MIMO in it is the abbreviation for virtual MIMO. The system performance greatly depends on deployment settings and system configurations below. In the simulation, all MSs are randomly dropped in a layout of three-tier 19 hexagonal cells with three identical sectors in each cell. The wraparound model is employed to simulate interference from neighboring cells. EESM is used to combine the SINRs on the subcarriers. Mobile WiMAX and LTE systems also have some specific key technologies. In Mobile WiMAX, synchronous HARQ with a four processes interlaced structure is adopted, and the maximum number of retransmissions is 4. The Chase Combing model is used to perform the SINR recalculation after the retransmission. Hard handover and AMC with six MCS levels are used in Mobile WiMAX. Besides, basic open-loop power control is applied in UL. How-
IEEE
BEMaGS F
1 Fairness criterion WiMAX LTE
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
0
0.5
1
1.5 2 2.5 Normalized throughput
3
3.5
4
Figure 3. Fairness curves of different systems in DL and UL.
ever, in LTE, the adaptive asynchronous HARQ, Intra-LTE handover, AMC with 15 MCS levels, and closed loop power control are adopted. The simulation platform is established by Visual Studio C++. The system-level simulation is based on the unified methodology with the above configurations. The simulation results about the throughput, spectrum efficiency, user fairness for Full Buffer traffic and capacity, packet loss, and user satisfaction rate for VoIP traffic are shown hereafter. Table 2 shows the performance at the peak user rate, the system throughput, and the spectrum efficiency for Full Buffer traffic in DL and UL. It is observed that the LTE system has higher system throughput and spectrum efficiency in both DL and UL. Moreover, the cell-edge performance in LTE is also better than that in Mobile WiMAX. Another performance index for Full Buffer traffic is the fairness among users, which can be achieved by using a proper scheduling algorithm, such as PF scheduler. The throughput and the spectral efficiency discussed above refer to aggregate system performance, whereas the fairness refers to the per-user performance, especially to the performance of cell-edge users. We should make an effective tradeoff between fairness and throughput. In order to achieve satisfying fairness with high spectral efficiency at the same time, we have calibrated the relevant parameters (_-factor in PF) in scheduling algorithms so that the following two criteria are met: • Fairness curves are similar to each other in shape and under the fairness criterion, which will guarantee the consistency and fairness among different systems. • Fairness curves are as close to the fairness criterion as possible, which indicates that the spectral efficiency is maximized. Figure 3 gives the user fairness curves in DL. It is shown that all curves are on the right side
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
CDF
Communications
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
153
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Simulation Parameters
Unified Value
System Configurations Frequency
2 GHz
Bandwidth
10 MHz
Duplex
TDD
DL:UL ratio
22:15 for WiMAX; 3:2 for LTE
Antenna configuration
DL:2 × 2; UL:1 × 2(or 2 × 2V-MIMO)
BS-to-BS distance
1 km
Minimum distance between BS and MS
35 m
Maximum UL total path loss
140 dB
Frequency reuse factor
1
Thermal noise density
–174 dBm/Hz
Scheduler
PF for Full Buffer; MLWDF for VoIP
Subchannel model for Mobile WiMAX
PUSC [3]
Propagation Parameters Propagation model
COST 231 Suburban
Log-normal Shadowing Std
8.9 dB
Correlation distance of shadowing
50 m
Shadowing correlation between cells
0.5
Shadowing correlation between sectors
1.0
Channel models
SCM-E
Penetration loss
20 dB
BTS Configurations BTS transmit power
43 dBm per antenna
BTS noise figure
5 dB
BTS antenna gain with cable loss
14 dBi
Antenna height of BS
30 m
Antenna horizontal pattern
70° with 20 dB front-to-back ratio
MS Configurations MS transmit power
23 dBm
MS noise figure
8 dB
MS antenna gain
–1 dBi
Antenna height of MS
1.5 m
MS velocity
3 km/h
MS number
10/sector
Traffic type
Full Buffer and VoIP
Table 1. Simulation parameters.
154
Communications IEEE
A
BEMaGS F
of the three-point fairness criterion curve, guaranteeing fairness among users. At the same time, each curve is as close to the fairness criterion as possible. The two curves are close to each other, which means that they are similar in terms of fairness. In addition, the probability of users with smaller throughput in LTE is lower than that in Mobile WiMAX, which accords with the cell-edge throughput in Table 2. The fairness curves in UL similar to that in Fig. 3 are omitted. Figure 4 gives the average ICI performance of LTE and Mobile WiMAX systems with respect to the number of users per sector. It shows that LTE has lower average ICI value than Mobile WiMAX, which means LTE can mitigate the ICI more effectively. With the increase of the user number, the average ICI of users decreases for both systems. Full Buffer is an ideal data traffic without a delay requirement, which can only reflect the throughput and the spectral efficiency performance. In the following, we evaluate the performance for VoIP traffic. The voice activity factor for VoIP is 50 percent, and the total voice payload on the air interface is 40 Bytes. The results for VoIP traffic in DL and UL are listed in Table 3. VoIP capacity is the average number of users in one sector when more than 95 percent of users satisfy the following three criteria: • The packet delay for each user is not more than 50ms with probability of 98 percent. • The packet loss for each user is lower than 3 percent. • Capacities for users in DL and UL are comparable. The user satisfaction rate is the proportion of users that satisfy the packet delay and the packet loss conditions. From the results, we can see that the LTE system has higher VoIP capacity than Mobile WiMAX with a little higher packet loss proportion in DL. Besides, the user satisfaction rate of LTE is nearly equal to or higher than that of Mobile WiMAX. Thus, the LTE system has better performance than the Mobile WiMAX system in general.
CONCLUSION In this article we proposed a general unified system-level simulation evaluation methodology for mobile cellular systems and presented the framework for the establishment of a unified simulation platform. The unified methodology highlights the features of air interfaces and the advantages of different system standards. Moreover, in order to compare various systems’ performance comprehensively, the unified modeling of various modules in different systems is studied. After that, the simulation structure and the general platform are proposed. Finally, based on the unified configurations and parameter settings, simulation results of LTE and Mobile WiMAX systems for both Full Buffer and VoIP traffics in DL/UL are presented through the proposed platform. The results show that the LTE system has higher spectrum efficiency, larger VoIP capacity, and better user satisfaction than
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Mobile WiMAX. The work in this article can serve as a reference for researchers, product developers, or engineers.
-73
BIOGRAPHIES LI CHEN [S] (
[email protected]) __________ received his B.Eng. degree in communication engineering from Beijing University of Posts and Telecommunications (BUPT) in 2007. Now he is pursuing his Ph.D. degree in communications and information systems at BUPT. His research interests focus on Network Coding, cross-layer optimization, radio resource management and performance evaluation of wireless communications. WENWEN CHEN (
[email protected]) ________________ received her B.Eng. (2007) degree in communication engineering, M.Eng. (2010) in communications and information systems from BUPT. She joined Qualcomm in 2010, as an engineer in Qualcomm Wireless Communication Technologies (China) Limited. Her research interests mainly focus on performance analysis of key technologies for B3G/4G systems. BIN WANG [S] (
[email protected]) _____________ received his B.Eng. degree in communication engineering from BUPT in 2009. Now he is pursuing his M.Eng. Degree in communications and information systems at BUPT. His research interests focus on Device to Device underlay cellular networks, cooperation communications, and performance evaluation of wireless communications. XIN ZHANG [M] (
[email protected]) _____________ received his B.Eng. (1997) degree in communication engineering, M.Eng. (2000) degree in signal and information processing, and Ph.D. (2003) degree in communications and information systems from BUPT. He joined BUPT in 2003, working in the Wireless Theories and Technologies Laboratory, and focuses his research mainly on key technologies and performance analysis of air interfaces for wireless networks. HONGYANG CHEN (
[email protected]) _________________ received his B.S. (2003) degree and M.S. (2006) degree in Institute of Mobile Communications from Southwest Jiaotong University, Chengdu, China. Currently, he is a Ph.D. student of the Graduate School of Information Science and Technology, University of Tokyo. In 2009, he was a visiting researcher in the UCLA Adaptive Systems Laboratory, under the supervision of Prof. Ali.H.Sayed. His research interests include Wireless Localization, Wireless Sensor Networks, Statistical Signal Processing. He has served as a TPC member for some flagship conferences. He organized a Session on WSNs at IEEE MILCOM’08. He received the Best Paper Award from IEEE PIMRC’09. He has been listed in Marquis Who’s Who in the World. D ACHENG Y ANG (
[email protected]) ____________ received his M.S. (1982) and Ph.D. (1988) degrees in circuits and systems
F
-75 Average ICI per user (dBm)
[1] 3GPP2 C.R1002-0 v.1.0, “cdma2000 Evaluation Methodology,” Dec. 2004. [2] R. Jain, S.-I. Chakchai, and A.-K. AL Tamimi, “SystemLevel Modeling of IEEE 802.16E Mobile WiMAX Networks: Key Issues,” IEEE Wireless Commun. Mag., vol.15, no.5, Oct. 2008, pp. 73–79. [3] WiMAX Forum, “WiMAX System Evaluation Methodology V4,” July 2008. [4] 3GPP TR 25.896, “Feasibility Study for Enhanced Uplink for UTRA FDD,” v.6.0.0, Mar. 2004. [5] 3GPP TS 25.214, “Physical Layer Procedures (FDD),” v.8.0.0, Dec. 2007. [6] Y. Gao et al., “Unified Simulation Evaluation for Mobile Broadband Technologies,” IEEE Wireless Commun. Mag., vol. 47, no. 3, Mar. 2009, pp. 142–49. [7] ITU-R Rec. M. 1225, “Guidelines for Evaluation of Radio Transmission Technologies for IMT-2000,” Jan. 1997. [8] 3GPP TR 25.996, “Spatial Channel Model for Multiple Input Multiple Output (MIMO) Simulations,” v.6.1.0, Sept. 2003. [9] L. Chen et al., “Inter-Cell Coordinated Resource Allocation for Mobile WiMAX System,” Proc. IEEE WCNC 2009, Budapest, Apr. 2009, pp. 1–6. [10] 3GPP TS 36.211, “Evolved Universal Terrestrial Radio Access (E-UTRA) Physical Channels and Modulation,” v.8.2.0, Mar. 2008.
-76 -77 -78 -79 -80 -81 -82
5
10
15
20 25 30 Number of users per sector
35
40
Figure 4. Average ICI per user with different user numbers.
UL 1 × 2 (2 × 2 V-MIMO)
DL Performance Indices WiMAX
LTE
WiMAX
LTE
Peak User Rate (Mb/s)
31.68
80.2
5.04 (5.04)
20.3 (20.3)
System Throughput (Mb/s)
7.91
8.23
2.38 (2.79)
2.74 (2.83)
Spectrum Efficiency (bps/Hz)
1.33
1.44
0.59 (0.69)
0.64 (0.66)
Cell-edge Throughput (Mb/s)
0.058
0.11
0.06 (0.04)
0.061 (0.03)
Table 2. Peak user rate, throughput and spectrum efficiency.
UL 1 × 2 (2 × 2 V-MIMO)
DL Performance Indices WiMAX
LTE
WiMAX
LTE
VoIP Capacity
240
390
128 (164)
250 (286)
User Average Throughput (kb/s)
9.8
8.3
7.2 (9.1)
7.95 (9.5)
Proportion of Users with Packet Loss > 3%
1.7%
2.0%
3.6% (3.6%)
2.4 (2.3)
User Satisfaction Rate in %
97.3
97.2
95.1 (95.9)
96.8 (97)
Table 3. Results for VoIP traffic. from BUPT. From 1992 through 1993 he worked at the University of Bristol, United Kingdom, as a senior visiting scholar, where he was engaged in Project Link-CDMA of the RACE program. In 1993 he returned to BUPT as an associate professor. Currently he is a professor at BUPT, working in the Wireless Theories and Technologies Laboratory, and his research interests are focused on wireless communications.
IEEE Communications Magazine • July 2011
IEEE
BEMaGS
LTE WiMAX
-74
REFERENCES
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
155
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
ACCEPTED FROM OPEN CALL
Layer 3 Wireless Mesh Networks: Mobility Management Issues Kenichi Mase, Graduate School of Science and Technology, Niigata University
ABSTRACT Wireless Mesh Networks (WMNs) may be broadly classified into two categories: Layer 2 and Layer 3 WMNs. This article focuses on the Layer 3 WMN, which provides the same service interfaces and functionalities to the conventional mobile host (MH) as the conventional wireless local area network. Three essential design issues to realize seamless mobility management in the Layer 3 WMN are identified and systematically discussed in this article. They are IP address resolution, location management, and Media Access Control (MAC) address resolution. The Layer 3 WMN backbone requires systematic management of the IP and MAC addresses of each MH, which can be realized by four basic approaches: centralized management, home management, replication management, and distributed management. It is shown that the address pair management architecture is fundamental to realizing efficient packet forwarding and address resolution. Design guidelines are provided to realize Layer 3 WMN supporting seamless MH roaming, considering the applicability of address pair management architectures with regard to WMN scale and client mobility; this applicability is considered based on a qualitative evaluation of the management overhead and handover performance.
INTRODUCTION Wireless communication has afforded remarkable convenience and benefits to human life, enabling communication at all times and from any place. Cellular phone networks and wireless local area networks (WLANs) are major vehicles to provide wireless communication. The former is categorized as a wide area network (WAN), while the latter is a local area network (LAN). WLANs can also be used to form access networks to WANs. The wireless mesh network (WMN) is an emerging technology to extend the use of wireless communication [1]. A WLAN backbone is composed of access points (APs) to accommodate mobile hosts (MHs) and typically wired LANs such as IEEE 802.3 to connect the APs. On the other hand, a WMN backbone is composed of mesh routers (MRs) instead of APs. MRs generally have a capability similar to APs in a WLAN for accommodating convention-
156
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
al MHs and are usually placed in a location similar to that of APs in a WLAN. However, unlike WLANs, neighboring MRs communicate with each other via a wireless link. If two MRs are not within direct wireless communication range, intermediate MRs may function to relay communication (i.e., wireless multihop communication). Typically, no wiring is necessary to construct the WMN backbone. This is a significant advantage of the WMN over the WLAN, which requires wiring between APs; hence, cost and time to deploy a network service is saved using the WMN. The WMN may be broadly classified into two categories: Layer 2 and Layer 3 WMNs. In the former, frame relaying (bridging) is performed in Layer 2 (the Media Access Control (MAC) layer) in the WMN backbone. A MAC address is used to deliver frames from one MH to another through the WMN backbone. Standardization of the Layer 2 WMN is under development in IEEE 802.11s. All wireless interfaces that form the WMN backbone require use of the IEEE 802.11s based devices. On the other hand, in the latter type of WMN, Internet Protocol (IP) packet relaying is performed in Layer 3 (Network layer) in the WMN backbone [2]. An IP address is used to deliver IP packets from one MH to another through the WMN backbone. There is no special requirement for the underlying wireless link systems used to connect the MRs. It is thus possible to select the most appropriate wireless link system such as IEEE 802.11a/b/g/n or IEEE 802.16 for each link in terms of cost and performance to form a heterogeneous wireless network for the WMN backbone; each MR in the WMN backbone may have multiple interfaces to different wireless link systems. The broadcasting of Address Resolution Protocol (ARP) request frames is necessary in the Layer 2 WMN backbone, consuming precious wireless bandwidth, while it can be avoided in the Layer 3 WMN backbone. For these reasons, Layer 3 WMN is expected to be more scalable than Layer 2 WMN. Various approaches and technical challenges to realize the Layer 3 WMN have been proposed and discussed in the literatures, though standardization is yet to begin. Mobility management is one of the major technical issues and challenges in the realization of the Layer 3 WMN. It is necessary for MHs to continuously send and/or
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
receive IP packets for the ongoing communication sessions during roaming in WMN. Readers may refer to [3, 4] for introductions and general discussions of mobility management in WMNs. This article focuses on mobility management for Layer 3 WMN that supports client side transparency for conventional MHs. The contributions of this article are: • The identification of the technical issues involved in designing Layer 3 WMN to support client side transparent mobility. • Discussing the common frameworks and architectures for solving these issues and realizing essential functions. • Presenting the design guidelines, including the selection of architectures.
MOBILITY MANAGEMENT ISSUES AND RELATED WORKS The architecture of WMNs can generally be classified into three types: Infrastructure/Backbone WMNs, Client WMNs, and Hybrid WMNs [1]. This article focuses on a primitive class of Infrastructure/Backbone WMNs, where MRs form a communication backbone for conventional IEEE 802.11 MHs to provide a service equivalent to that of WLAN (Fig. 1). Hereafter, this type of WMN is referred to as “Layer 3 WMN” or simply “WMN.” An MH establishes a linklayer association with one of the MRs through IEEE 802.11 infrastructure mode protocol and obtains an IP address through a Dynamic Host Configuration Protocol (DHCP) service provided in the WMN, when it first enters the WMN. It may roam within the service coverage of WMN. The link-layer handover from one MR to another follows the IEEE 802.11 handover procedure. Instead of the IEEE 802.11 infrastructure mode, an ad hoc mode is used both for MRs and MHs in [5, 6]. However, such a restriction is not appropriate for meeting our objective of realizing a WMN that can be used instead of WLAN and is hence out of the scope of this article. There are two approaches to supporting mobility management for the Layer 3 WMN. In the first approach, each MH has two IP addresses, one for representing MH identification and the other for representing the location of the MH, which is the current point of attachment of the MH to the WMN (one of MRs), as in the Mobile IP (two-tier addressing model). Each MR configures an individual Extended Service Set (ESS) composed of one Basic Service Set (BSS). When an MH roams from one MR to another in the same WMN, Layer 2 handover occurs. In addition, it conducts a new DHCP request and response to configure its IP address for representing the location of MH (care-ofaddress). In the second approach, each MH has a single IP address. All MRs cooperatively configure a single ESS. MHs can maintain their IP address constant (single addressing model) during MR-to-MR roaming. Just a link-layer handover is sufficient for MHs to realize seamless roaming. Since IP packet forwarding is performed in the Layer 3 WMN backbone, the two-tier
The Internet
MR
IEEE
BEMaGS
Heterogenous wireless network
GW MR
F
MR
Wireless link
MR
MR
MR: Mesh router GW: Gateway MH: Mobile host
IEEE802.11 WLAN MH MH MH
MH MH
MH MH MH
Figure 1. An example of wireless mesh network structures.
addressing model is quite straightforward, where MHs need to have functionalities to support the Layer 3 handover [7–9]. On the other hand, in the single addressing model, no special functionalities are required for MHs (client side transparency in [4]). Since our goal is to provide the same backbone service for conventional MHs as in the WMN, only the client side transparency seems to be a promising approach. However, in this approach, special care is required for IP packet forwarding without any dependence on MH functionalities [10–13]. There are three essential design issues for supporting mobility management for a single addressing model. IP address resolution: Since an IP address is used to deliver IP packets from one MH to another through the Layer 3 WMN backbone, each MR needs to recognize the IP addresses of the MHs that are associated to it. When an MH roams from one MR to another, it completes a link-layer association with a new MR. As a result, the MR can recognize the MAC address of the MH. However, it cannot freely obtain the IP address of the MH. In [14], DHCP servers on MRs assign IP addresses to the MHs based on the MAC addresses of the MHs, so that the MR can infer the IP address of the roamed MH from its MAC address. However, this prevents efficient IP address assignment, for instance, the possible use of address aggregation, as mentioned in the next section. How then is the IP address of the roamed MH obtained? Location management: Since each MH has only one IP address in the single addressing model and possibly roams from one MR to another in the Layer 3 WMN backbone, this IP address represents MH identification but does not represent its location, unlike the care-ofaddress used in Mobile IP. How is the location of an MH to which the packets are delivered identified? In MobileNAT, the address representing MH identification is translated into the address representing the current point of the MH attachment to the Internet and is used for routing packets to the MH [7, 8]. This technique can be used when an MH only communicates with a corresponding node in the Internet but cannot be used for communications within the WMN. On the other hand, a routing protocol operating in the WMN backbone can be used to create and maintain the routing table of each MR, which includes the IP addresses of the MHs
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
157
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
5: Route update AMS
3: Response
5: Route update
1: Roaming
4
2: IP address inquiry
4: IP address notification
7: Packet forwarding
6: Packet sending
Figure 2. Packet forwarding mechanism (centralized management-Scheme A).
as the destination (flat routing or host routing). The roaming of MHs updates the routing table for the related MRs, resulting in a possible high message overhead consuming wireless bandwidth. How then is efficient location management conducted? MAC address resolution: Since Layer 3 WMN provides one ESS for MHs, ARP needs to be supported for it. An MH (source MH) of one MR may send an ARP request to another MH (destination MH) belonging to a different MR. For a straightforward emulation of ARP in the Layer 3 WMN backbone, a flooding ARP request over all the MRs of the WMN backbone is required [12]; this ARP request consumes wireless bandwidth and should be avoided. In [13], a kind of proxy ARP was proposed. Upon receiving the ARP request, the MR replies to the ARP request from the MH with its own MAC address. In [14], the MR replies with a fake MAC address that is uniform on all MRs, and hence, the source MH does not need to update its ARP cache when it roams to another MR. Unfortunately, in both approaches, a problem arises when the source and destination MHs meet in the same MR later. How then is efficient ARP performed? Several studies have been conducted on mobility management in WMNs. However, the three design issues mentioned above have not yet been fully discussed. This article presents frameworks to systematically address these issues.
158
Communications IEEE
A
BEMaGS F
assignment for MHs in cooperation with the DHCP relay servers located in the MRs. In type II, the address block is divided into address block units with a pre-determined address range. Each MR requests an address block unit to the address management server (AMS) in the WMN backbone. Upon receiving the request, the AMS assigns one or more address block units in the form of an address prefix to the requesting MR. An MH is assigned an IP address from the range of the address block units provided to the MR it associates with. (This MR is hereafter referred to as the “home MR.”) Each MR may have a DHCP server to perform efficient DHCP service, although a single DHCP server arrangement is also possible. An MH may roam from one MR to another, but it is expected that the majority of MHs remain in the original location (home MR). For these MHs, the prefixes of their IP addresses represent their locations. Two MHs in the same Layer 3 WMN use the MAC address to identify the sender and receiver of the frame even if they connect different MRs, while the MRs in the WMN backbone use the IP address to identify the sender and receiver of the corresponding IP packet included in the frame in order to forward the IP packet. To perform packet forwarding in the WMN backbone, the MAC and IP addresses (an address pair) of each MH and its location (the current MR) need to be systematically managed in the Layer 3 WMN. This article presents four address pair management architectures, the first three architectures further classified into Schemes A and B, to provide a framework for the mapping between the MAC address and the IP address of each MH. This mapping is used to solve two design issues — IP address resolution and MAC address resolution — mentioned in the previous section. The address pair management architecture thus provides a foundation for the packet forwarding mechanism based on the IP address resolution (see next section) and MAC address resolution mechanism (see later sections).
CENTRALIZED MANAGEMENT
ADDRESS PAIR MANAGEMENT ARCHITECTURES
In Scheme A, a centralized server maintains the address pair records of all MHs in the WMN [13]. This server is also referred to as AMS for convenience, but it may be different from the AMS used for IP address assignment. In Scheme B, the AMS additionally maintains the current MR records of all MHs. When an IP address is assigned to an MH, the home MR obtains the address pair and sends it to the AMS. When an MH roams between MRs, the current MR obtains the MAC address of the MH and sends it to the AMS, which returns the IP address of the MH to the current MR (see 1-3 in Fig. 2). In addition, in Scheme B the AMS updates the current MR of the MH (see 1-3 in Fig. 3).
An MH needs to obtain an IP address when it newly joins the WMN, that is, when it establishes Layer 2 association with one of the MRs of the WMN. Two types of the DHCP service can be considered. In type I, an address block reserved for MHs is provided for the entire WMN and each MH is assigned an IP address from the range of this address block. A DHCP server located in the WMN performs IP address
In Scheme A, each MR maintains the address pair records of the MHs for which it is the home MR. In addition, a centralized server maintains the MAC address and the home MR pair records of all MHs in the WMN. This server is termed the home server (HS). In Scheme B, each MR additionally maintains the current MRs of the
HOME MANAGEMENT
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
MHs for which it is the home MR. When an IP address is assigned to an MH, the home MR reports the MAC address and the home MR (itself) of the MH to the HS. When an MH roams between MRs, the current MR obtains the MAC address of the MH and sends it to the HS, which returns the home MR of the MH to the current MR. The current MR informs the home MR of the MAC address of the MH and obtains the IP address of the MH in return. In addition, in Scheme B the home MR updates the current MR of the MH [11].
REPLICATION MANAGEMENT In Scheme A, each MR maintains the address pair records of all MHs in the WMN [10, 11]. In Scheme B, each MR additionally maintains the current MRs of all MHs. When the IP address is assigned to an MH, the home MR obtains the address pair and sends it to all the other MRs using flooding or multicast protocols. When an MH roams between MRs, the current MR obtains the MAC and IP addresses of the MH using its own address pair records. In Scheme B, when an MH roams between MRs, the current MR obtains the MAC address of the MH and sends it (or its IP address) to all the other MRs to notify the current MR of the MH and each MR updates the current MR of the MH (see 1-3 in Fig. 4).
DISTRIBUTED MANAGEMENT Each MR maintains the address pair records of MHs with which it has or had an association [15]. When an MH roams between MRs, the current MR obtains the MAC address of the MH. If the current MR does not know the corresponding IP address, it sends an IP address request that includes the MAC address of the MH to the neighboring MRs based on the expanded ring search. In this ring search, the maximum number of searches is given and the maximum number of hops per search increases with each search trial [15]. The neighboring MR that maintains the address pair record of the requested MH returns the IP address of the MH to the requesting MR (see 1–3 in Fig. 5). Assuming the MH’s typical roaming speed from one MR to another, it is highly probable that one of the one-hop neighboring MRs has the requested address pair record. The maximum number of hops in the first search may thus be limited to one or two hops, and the first search is expected to almost always succeed.
PACKET FORWARDING MECHANISM Consider IP packet forwarding between MHs in a Layer 3 WMN backbone. The MH that sends the packets and its current MR are termed the source MH and source MR, respectively. The MH that receives the packets and its current MR are termed the destination MH and destination MR, respectively. The source and destination MHs may be accommodated in the same MR or in different MRs. In the former case, each frame that conveys an IP packet is relayed to the destination MH at Layer 2 using the AP capability of the MR, as is usual in WLAN. In the latter case, packet forwarding is performed
IEEE
BEMaGS F
AMS 3: Current MR update 2: IP address inquiry
3: Current MR inquiry
3: Response
6: Response 1: Roaming 7: Packet forwarding
4: Packet sending
Figure 3. Packet forwarding mechanism (centralized management-Scheme B). in the manner of either flat or hierarchical routing. Flat routing is employed in centralized management-Scheme A, home management-Scheme A, replication management-Scheme A, and distributed management. MRs have route entries for destination MHs in their routing table. Efficient location management in these schemes is one of the design issues mentioned in the second section. Hierarchical routing is employed in centralized management-scheme B, home management-Scheme B, and replication managementScheme B. The source MR first identifies the destination MR of the destination MH and then sends the packets to the destination MR by means of IP-in-IP encapsulation or IPv6 routing header (tunneling). The destination MR then forwards the received packets to the destination MH. MRs do not have the route information for destination MHs but for destination MRs in their routing table. In both flat and hierarchical routing, routing protocol runs in the WMN backbone. In flat routing, MH information (IP address) is explicitly included in the routing messages for route creation and update, while in hierarchical routing, it is not. The mobile ad hoc network (MANET) routing protocol is a reasonable candidate to be used for the WMN backbone, since it supports the wireless multihop packet forwarding capability. MANET routing protocols are generally classified into proactive and reactive routing protocols. Either of these can be used in the WMN backbone. The choice of the appropriate routing depends on the scale of the WMN, traffic characteristics, and other conditions. Next, the packet forwarding mechanism for each of the address pair management approaches is described.
CENTRALIZED MANAGEMENT Scheme A — Each MR periodically notifies the IP address information of the MHs that associate with it to other MRs in routing messages (proactive routing) or in response to the route request (reactive routing). If IP address assignment type II is used in proactive routing, it does
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
159
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
MR from the IP address of the destination MH, IP address assignment-type II is assumed above. A kind of home management-Scheme B-Method 2 is termed the transparent MIP (Mobile IP) in [11].
3: Current MR update
REPLICATION MANAGEMENT Scheme A — The route update is performed in a manner similar to that in centralized management-Scheme A.
1: Roaming 3: Current MR update
2 5: Packet forwarding
2: Current MR notification
4: Packet sending
Scheme B — Each MR can calculate or discover a route to other MRs by means of a routing protocol that runs in the WMN backbone. The source MR owns the record of the current MR of the destination MH and forwards the packet to the current MR through tunneling (see 4–5 in Fig. 4).
DISTRIBUTED MANAGEMENT Figure 4. Packet forwarding mechanism (replication management-Scheme B).
not have to advertise the individual IP address information of the associating MHs for which it is the home MR. Instead, it advertises the prefix information of the address block units that it owns. As a result, the amount of IP address information of the MHs included in the routing messages is substantially reduced to perform efficient location management. A source MR can then calculate or discover a route to the AMS and the destination MH by means of a routing protocol that runs in the WMN backbone and forwards the packet to the destination MH (see 4–7 in Fig. 2). Scheme B — Each MR can calculate or discover a route to AMS and other MRs by means of a routing protocol that runs in the WMN backbone. If the source MR does not hold the current MR of a received packet, it sends the location request that includes the IP address of the destination MH to the AMS, which then returns the current MR of the destination MH. The source MR then forwards the packet to the current MR through tunneling (see 4–7 in Fig. 3).
HOME MANAGEMENT Scheme A — The route update is performed in a manner similar to that in centralized management-Scheme A. Scheme B — Each MR can calculate or discover a route to other MRs by means of a routing protocol that runs in the WMN backbone. In Method 1, if the source MR does not hold the current MR of a received packet, it sends the location request that includes the IP address of the destination MH to its home MR, which returns the current MR of the destination MH. The source MR then forwards the packets to the current MR through tunneling. In Method 2, the source MR sends packets to the home MR of the destination MH, after which the home MR forwards the packets to the current MR; both the sending and forwarding of the packets employ tunneling. To directly resolve the home
160
Communications IEEE
The route update is performed in a manner similar to that in centralized management-Scheme A (see 4-6 in Fig. 5).
MAC ADDRESS RESOLUTION MECHANISM In WLAN, a source MH broadcasts an ARP request that includes its MAC and IP addresses and the IP address of the destination MH, when the MAC address of the destination MH does not exist in its ARP table. Upon receiving the ARP request, the destination MH sends back an ARP reply, which includes its MAC and IP addresses, to the source MH and updates its ARP table with regard to the source MH. The source MH updates its ARP table when it receives the ARP reply. Similarly, ARP needs to be supported in the Layer 3 WMN to provide the ESS service to MHs. The receiver MAC address of the frame originating from the source MH and conveying an IP packet to be sent to the destination MH is thus set to the MAC address of the destination MH. Consider the case that the source and destination MRs are different. In such a case, IP packets received from the source MH are forwarded from the source MR to the destination MR without using the MAC address of the destination MH set in the receiver MAC address of the frames originating from the source MH. Thus, one may consider that faithful (correct) address resolution is not necessary and that instead of returning the actual MAC address of the destination MH, a dummy MAC address is returned by the source MR [13, 14]. However, unfortunately, this consideration is inaccurate, because if the source MH and destination MH later meet in the same MR, the frames cannot be relayed (bridged) in the Layer 2 of the MR; therefore, Layer 3 forwarding is required, which unnecessarily increases the Layer 3 processing load of the MR. Next, the ARP mechanism for all the centralized and distributed management approaches is described under the assumption that the route information to the destination MH or the destination MR is proactively calculated or reactively discovered. Due to space constraints, the ARP
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
mechanism for other approaches is not discussed. The proposed ARP mechanism performs faithful ARP replies, while the ARP request multicast is not necessary in the WMN backbone.
F
5: Route update 3: Response
CENTRALIZED MANAGEMENT Scheme A — When a source MR receives an ARP request and contains the reply information (MAC address of the destination MH) in its ARP table, it returns the ARP reply on behalf of the destination MH. Otherwise, it sends an ARP request packet (IP packet that includes the ARP request message) to the AMS. Upon receiving this request, the AMS returns an ARP reply packet (IP packet that includes the ARP reply message) to the source MR. When the source MR receives this reply, it updates its ARP table and returns the ARP reply to the source MH on behalf of the destination MH. It also sends a gratuitous ARP reply packet (IP packet that includes the gratuitous APR message) to the destination MH. The destination MR intercepts this ARP reply packet, updates its ARP table, and sends the gratuitous ARP reply to the destination MH on behalf of the source MH. The source and destination MHs update their ARP tables when they receive the ARP and gratuitous ARP replies, respectively. Scheme B — The procedure is the same as that employed in Scheme A. The difference is in the sender and receiver of the gratuitous ARP. In Scheme A, the source MR sends a gratuitous ARP reply packet to the destination MH, while in Scheme B, the AMS sends a gratuitous ARP reply packet to the destination MR, since the AMS has the current MR record of the destination MH. Upon receiving the ARP reply, the destination MR updates its ARP table and sends the gratuitous ARP reply to the destination MH on behalf of the source MH.
DISTRIBUTED MANAGEMENT When a source MR receives an ARP request and it contains the information in its ARP table, it returns the ARP reply on behalf of the destination MH; otherwise, it sends an ARP request packet to the destination MH. The destination MR intercepts this ARP request, updates its ARP table, and returns an ARP reply packet to the source MR. Upon receiving this reply, the source MR updates its ARP table and returns the ARP reply to the source MH on behalf of the destination MH. The destination MR also sends a gratuitous ARP reply to the destination MH on behalf of the source MH. The source and destination MHs update their ARP tables when they receive the ARP and gratuitous ARP replies, respectively. It should be noted that this method can also be employed in centralized management-Scheme A and home managementScheme A.
DESIGN GUIDELINES The mechanisms and features of address pair management architectures for realizing Layer 3 WMN that supports client side transparent mobility management are summarized in Table
2: IP address inquiry 1: Roaming 5: Route update
4 Packet forwarding
4: IP address notification
6: Packet sending
Figure 5. Packet forwarding mechanism (distributed management).
1. The requirements and guidelines in designing and selecting the Layer 3 WMN architectures with consideration of their applicability to WMN scale and client mobility are given below.
BASIC REQUIREMENT The mobility management architectures should satisfy the following two requirements: • Data frames that are sent from one MH to another belonging to the same MR should be relayed in Layer 2. Note that this condition is always satisfied in the conventional WLAN, where AP is used to bridge data frames in the BSS. If Layer 3 processing is required to relay data frames within the MR of WMN, the MR requires higher computing power than the AP of the WLAN and may act as a bottleneck to processing. • ARP request message flooding or multicast should be avoided to preserve the precious wireless bandwidth of the WMN. All of the four basic architectures presented in this article satisfy the requirements mentioned above and recommended as the basic architecture options in designing Layer 3 WMN.
ARP MESSAGE OVERHEAD The ARP message overhead within each BSS is common in all architectures. In the WMN backbone, the ARP request and reply messages require two-way unicast message propagation, while the gratuitous ARP reply requires oneway unicast message propagation; the former message overhead is approximately double the latter one. The relative cost for the ARP message overhead is shown in Table 1, assuming that the number of hops from the source MR to the AMS, to the home MR, or to the destination MR is the same, and that the two-way unicast message propagation cost is 1 unit. Since the ARP message traffic is usually smaller than the data packet traffic, the ARP cost difference between different architectures may not be the primary factor for selecting the architecture.
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
161
A
BEMaGS F
Communications
Replication management
Home management
Centralized management
IEEE
Scheme A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Address pair location
Current MR location
AMS
N/A
Address pair & Current MR management message overhead Joining
Roaming
IP address inquiry to AMS Address pair to AMS
AMS
IP address inquiry to AMS Current MR notification
Current MR inquiry to AMS Tunneling
Scheme A
Home MR
N/A
Home MR inquiry to HS IP address inquiry to home MR
Route update
Home MR inquiry to HS IP address inquiry to home MR Current MR notification to home MR
Current MR inquiry to home MR (method 1) Tunneling (Method 1 & Method 2)
N/A
Route update
Home MR
Home MR
Scheme A
Each MR
N/A
Scheme B
Distributed management
Address pair to all MRs Each MR
Each MR
Current and past MRs
N/A
N/A
ARP request & reply (Relative cost)
Route update
AMS
Scheme B
BEMaGS F
ARP message overhead Routing message overhead
Scheme B
MAC address & home MR to HS
A
Forwarding increase
Handover delay
From source MR to destination MH (0.5)
N/A
IP address inquiry to AMS Route update
From AMS to destination MR (0.5)
Current MR inquiry to AMS
IP address inquiry to AMS
From source MR to destination MH (0.5)
N/A
Home MR inquiry to HS IP address inquiry to home MR Route update
From home MR to destination MR (0.5)
Current MR inquiry to home MR (Method 1) Triangular routing (Method 2)
Home MR inquiry to HS IP address inquiry to home MR Current MR notification to home MR
Gratuitous ARP reply
From source MR to AMS (1)
From source MR to Home MR (1)
From source MR to destination MH (0.5) N/A
Current MR notification to all MRs
Tunneling
IP address inquiry to neighbor MRs
Route update
Route update N/A
From source MR to destination MR (0.5) From source MR to destination MH (1)
N/A
Current MR notification to all MRs
N/A
IP address inquiry to neighbor MRs Route update
Table 1. Comparison between address pair management architectures.
CENTRALIZED MANAGEMENT VERSUS HOME MANAGEMENT Among the four basic architecture approaches, centralized management and home management are similar since both these approaches require a centralized sever, AMS, and HS, respectively. Two successive inquiries for resolving the home MR and IP address are required in the case of roaming in home management, which is its main drawback as compared to centralized management. Home management may thus be excluded from the architecture options.
SCHEME B OPTION IN MANAGEMENT ARCHITECTURES In Scheme B, the current MR of each MH is additionally maintained in the AMS in case of centralized management and in all the MRs in case of replication management. In centralized management-Scheme B, current MR notification is performed together with IP address inquiry to the AMS, which is also necessary in Scheme A. Therefore, Scheme B is almost the same as Scheme A in terms of message overhead. On the other hand, in replication management, IP address inquiry is not necessary and current MR notification needs to be independently provided for Scheme B. Therefore, Scheme B is inferior
162
Communications IEEE
to Scheme A in terms of message overhead, and the advantage of Scheme B over Scheme A is debatable. Replication management-Scheme B may thus be excluded from the architecture options.
SCHEME A VERSUS SCHEME B IN CENTRALIZED MANAGEMENT In Scheme A, routing overhead and handover delay significantly depend on inter-MR mobility (the number of MHs that roam from one MR to another with respect to time). With high interMR mobility, route update needs to be frequently performed to quickly follow the change in MH locations, thereby increasing the routing overhead. Otherwise, the route update is delayed, resulting in increasing handover delay. On the other hand, in Scheme B, routing overhead and handover delay do not depend on inter-MR mobility, since route update is not necessary. Therefore, Scheme B can be superior to Scheme A when inter-MR mobility is significantly high.
ADDRESS PAIR MANAGEMENT OVERHEAD The address pair management overhead significantly depends on the MH changing rate (the number of new MHs joining WMN with time). Consider the following two cases:
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Case 1 — Low MH changing rate: Roaming overhead is of primary concern since joining overhead is not significant. In replication management-Scheme A the roaming overhead is zero, while in other managements, it increases with inter-MR mobility. Replication management-Scheme A is thus an attractive choice in the case of high inter-MR mobility. However, the joining overhead may still become significant when the WMN scale is quite high in replication management-Scheme A. In such a case, distributed management may also be considered a suitable option. Case 2 — High MH changing rate: The joining overhead is of considerable concern. It is zero in case of distributed management, while it increases with MH changing rate in case of other management approaches. In distributed management, the roaming overhead remains almost unchanged with the WMN scale, since the IP address inquiry is targeted only at the neighbor MRs. Distributed management is thus superior to other management options.
CONCLUSIONS Wireless mesh network (WMN) is an emerging mobile network technology, and extensive studies are being conducted to realize scalable and high-performance WMNs. This article identified the essential technical issues in designing Layer 3 WMN that supports client side transparent mobility management. In this design, the Layer 2 system for MHs and the Layer 3 system for MRs need to be seamlessly combined in MRs. The most fundamental issue in terms of architecture is to efficiently maintain the address pair (MAC and IP addresses) and the location information of each MH that newly joins and roams in a WMN. Four address pair management architectures — centralized management, home management, replication management, and distributed management — are presented. The first three management architectures are further classified into Scheme A and Scheme B, which employ flat and hierarchical routing, respectively. This article discusses the use of these management architectures as the frameworks for realizing the essential mechanisms of the Layer 3 WMN — packet forwarding and address resolution. Finally, the requirements and guidelines in designing and selecting Layer 3 WMN architectures supporting client side transparent mobility management are given, considering their applicability to WMN scale and client mobility based on the qualitative evaluation of message overhead and handover performance. This article can be used to lay a foundation and provide frameworks for designing and developing the practical Layer 3 WMN.
IEEE
BEMaGS F
REFERENCES [1] I. F. Akyildiz, X. Wang, and W. Wang, “Wireless Mesh Networks: A Survey,” Computer Networks, vol. 47, no. 4, pp. 445–87, 2005. [2] K. Mase et al., “A Testbed-based Approach to Develop Layer 3 Wireless Mesh Network Protocols,” Tridentcom2008. [3] J. Xie and X. Wang, “A Survey of Mobility Management in Hybrid Wireless Mesh Network,” IEEE Network, vol. 22, no. 6, 2008, pp. 34–40. [4] V. Mirchandani and A. Prodan, “Mobility Management in Wireless Mesh Networks,” Comp. Commun. and Networks, 2009, pp. 349–78. [5] S. Speicher, “OLSR-FastSync: Fast Post-Handovers Route Discovery in Wireless Mesh Networks,” Proc. IEEE Vehic. Tech. Conf. (VTC’06-Fall), 2006, pp. 1–5. [6] Y. Amir et al., “Fast Handover for Seamless Wireless Mesh Networks,” Proc. ACM Int’l. Conf. Mobile Syst., Apps., Services (MobiSys), 2006, pp. 83–95. [7] M. Buddhikot et al., “MobileNAT: A New Technique for Mobility Across Heterogeneous Address Spaces,” ACM Mobile Networks and Applications, vol. 10, no. 3, 2005, pp. 289–302. [8] K. N. Ramacandran et al., “On the Design and Implementation of Infrastructure Mesh Networks,” Proc. 1st IEEE Wksp. Wireless Mesh Networks, 2005. [9] A. M. Srivatsa and J. Xie, “A Performance Study of Mobile Handover Delay in IEEE 802.11-Based Wireless Mesh Networks,” Proc. IEEE Int’l. Conf. Commun. (ICC), 2008, pp. 2485–89. [10] Y. Owada and K. Mase, “A Study on Protocol, Implementation and Throughput Evaluation for Multihop Wireless LAN,” IEEE Vehic. Tech. Conf. (VTC 2003Spring), vol. 3, 2003, pp. 1773–77. [11] V. Navda, A. Kashyap, and S. R. Das, “Design and Evaluation of iMesh: An Infrastructure-Mode Wireless Mesh Network,” IEEE Int’l. Symp. A World of Wireless, Mobile and Multimedia Networks (WoWMoM), 2005, pp. 164–70. [12] L. Iannone and S. Fdida, “MeshDV: A Distance Vector Mobility-Tolerant Routing Protocol for Wireless Mesh Networks,” IEEE ICPS Wksp. Multi-hop Ad Hoc Networks: From Theory to Reality (REALMAN), 2005. [13] H. Wang et al., “A Network-Based Local Mobility Management Scheme for Wireless Mesh Networks,” Proc. IEEE Wireless Commun. Net. Conf. (WCNC), 2007, pp. 3795–800. [14] M. Ren et al., “MEMO: An Applied Wireless Mesh Network with Client Support and Mobility Management,” Proc. IEEE Global Telecom. Conf. (GLOBECOM), 2007, pp. 5075–79. [15] N. Azuma, K. Mase, and H. Okada, “A Proposal of Low-Overhead Routing Scheme for Layer 3 Wireless Mesh Networks,” Int’l. Symp. Wireless Pers. Multimedia Commun., 2009.
The most fundamental issue in terms of architecture is to efficiently maintain the address pair (MAC and IP addresses) and the location information of each MH that newly joins and roams in a WMN.
BIOGRAPHY KENICHI MASE [F] (
[email protected]) _____________ received the B. E., M. E., and Dr. Eng. Degrees in Electrical Engineering from Waseda University, Tokyo, Japan, in 1970, 1972, and 1983, respectively. He joined Musashino Electrical Communication Laboratories of NTT Public Corporation in 1972. He was Executive Manager, Communications Quality Laboratory, NTT Telecommunications Networks Laboratories from 1994 to 1996 and Communications Assessment Laboratory, NTT Multimedia Networks Laboratories from 1996 to 1998. He moved to Niigata University in 1999 and is now Professor, Graduate School of Science and Technology, Niigata University, Niigata, Japan. He received IEICE best paper award for the year of 1993 and the Telecommunications Advanced Foundation award in 1998. His research interests include communications network design and traffic control, quality of service, mobile ad hoc networks and wireless mesh networks. He is an IEICE Fellow.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
163
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
ACCEPTED FROM OPEN CALL
Preamble Design, System Acquisition, and Determination in Modern OFDMA Cellular Communications: An Overview Michael Mao Wang, Avneesh Agrawal, Aamod Khandekar, and Sandeep Aedudodla
ABSTRACT The wide choices of deployment parameters in next generation wireless communication systems, such as flexible bandwidth allocation, synchronous/asynchronous modes, FDD/TDD, full/half duplex, and configurable cyclic prefix duration, etc., present significant challenges in preamble and system acquisition design. This article addresses these challenges, as well as the solutions provided by the next generation wireless standards, such as the IEEE 802.20 Mobile Broadband Wireless Access (MBWA) standard and the 3GPP LTE standard. The design facilitates the maximal flexibility of the system configuration and yet has low overhead, low acquisition latency, and low complexity. System acquisition and determination techniques are also discussed in detail.
INTRODUCTION In a wireless communication system, an access terminal must acquire the system information before it can access the network. A preamble is a special signal for system acquisition, i.e., to provide a means for the access terminal to acquire the system parameters that are necessary to access the system. That is, the preamble serves as a “gateway” for terminals to gain access to the network. A wireless network periodically transmits a preamble signal for terminals to acquire. The goals of the preamble design are low overhead and low acquisition complexity. However, these two goals can be hard to achieve simutaneously for a highly flexible communication system as high flexibility typically incurs high overhead and/or high acquisition complexity. Flexible configuration for variable deployment requirements is one of the important and highly desirable features for the next generation cellular communication systems (or wireless wide area networks). However, the wide choices of deployment parameters and modes present significant challenges in acquisition system design. This article addresses these challenges and provides the design solution that has been adopted by the IEEE 802.20 Mobile
164
Communications IEEE
0163-6804/11/$25.00 © 2011 IEEE
Broadband Wireless Access standard (MBWA) [1, 2] as well as other emerging wireless WAN standards [3–5] such as 3GPP LTE. Without loss of generality, this article uses IEEE MBWA as a design paradigm. The IEEE MBWA standard is the new high mobility standard developed by the 802.20 Working Group of the 802 Committee [1], to enable high-speed, reliable, cost-effective mobile wireless connectivity. IEEE MBWA is a standard optimized to provid IP-based broadband wireless access in a mobile environment. MBWA can operate in a wide range of deployments, thereby affording network operators with superior flexibility in optimizing their networks. This standard is targeted for use in a wide variety of licensed frequency bands and regulatory environments. For example, MBWA can operate in a wide range of bandwidths (2.5–20 MHz); this flexibility enables an operator to customize a MBWA system for the spectrum available to the operator. MBWA has a unified design for full and half duplex FDD and TDD and a scalable bandwidth from 2.5 to 20 MHz for variable deployment spectrum needs. Similar to MBWA, LTE supports system bandwidth from 1.4 to 20 MHz. IEEE MBWA employs orthogonal frequency division multiple access (OFDMA) technology, which utilizes the orthogonal frequency division multiplexing (OFDM) as its key modulation technique [6]. The subcarrier spacing is fixed at 9.6 kHz corresponding to an OFDM symbol duration of Ts 5 104+s. The length of the cyclic prefix of an OFDM symbol is variable, T CP = NCPTS/16 5 6.51NCP +sec, where NCP 5 1, 2, 3, 4. This allows the operator to choose a cyclic prefix length that is best suited to the expected delay spreads in the deployment. At an MBWA transmitter, the transmitted data are organized as superframes. For a MBWA access network, a superframe consists of a preamble followed by 25 PHY frames in the FDD mode or 12 PHY frames in TDD mode. Both the preamble and the PHY frames consist of eight OFDM symbols. The preamble is used by an access terminal for the purpose of system acquisition while the PHY frames are used for
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
PHY frame (forward/ reverse link)
H
1
2
3
4
Acquisition pilot 3
0
Acquisition pilot 2
OFDM symbol index
5
6
7
IEEE
BEMaGS F
Figure 1. Superframe and preamble structure in FDD/TDD mode.
PREAMBLE STRUCTURE The IEEE MBWA preamble consists of eight OFDM symbols. The first OFDM symbol is used to transmit the Primary Broadcast Control Channel (PBCCH), while the next four OFDM symbols are used to transmit the SBCCH (Secondary Broadcast Control Channel). The remaining three OFDM symbols carry Acquisition Pilots 1, 2, and 3. The structure of the superframe preamble is depicted in Fig. 1. The ordering of the preamble OFDM symbols, i.e., placing five PBCCH/SBCCH OFDM symbols before the Acquisition Pilots, is to provide sufficient receiver AGC (automatic gain control) convergence time for the Acquisition Pilots during initial acquisition since Acquisition Pilots are acquired first as will be seen later. As shown in Fig. 2, the received preamble power can be very different from that of the PHY frames. For example, in the TDD mode, the PHY frame before the preamble belongs to the reverse link; the receive power can be either very high or very low depending on the position of the nearby terminals. Therefore, the receiver AGC gain at the last PHY frame of the superframe can differ greatly from the gain required for sampling Acquisition Pilot 1. Hence, the receiver AGC may require a large amount of time to converge to the right gain for acquiring Acquisition Pilot 1. The five OFDM symbols in front of the Acquisition Pilot 1 provide sufficient convergence time for AGC. Once Acquisition Pilots are acquired, the same AGC gain for the Acquisition Pilots can be set for the sampling of PBCCH and SBCCH OFDM symbols. One shortcoming of this design is the increased acquisition delay. After Acquisition Pilots are acquired, the receiver has to wait till the next preamble in order to decode the PBCCH and SBCCH packets unless the whole preamble (eight OFDM symbols) are buffered. Another shortcoming is that in order for the receiver to locate the next preamble, further information (e.g., FDD/TDD mode, etc) that affects the superframe length is necessary to be embedded in the Acquisition Pilots. The preamble transmission is limited to the
IEEE Communications Magazine • July 2011
Communications
A
PHY frame (forward/ reverse link)
PHY frame (forward link)
Acquisition pilot 1
Preamble (forward link)
CC
data traffic transmission. In FDD half duplex mode, each PHY frame is separated by a guard interval (3Ts/4), whereas there is no separation in full duplex mode. In TDD mode, one or two PHY frames are transmitted continuously on the forward link and one PHY frame is transmitted continuously on the reverse link, resulting in a 1:1 or 2:1 partitioning. In a typical 1:1 partitioning, the 12 forward and the 12 reverse PHY frames are interlaced and separated by guard intervals, 3Ts/4 between a forward and a reverse PHY frame, and 5Ts/32 between a reverse and a forward frame [1]. It is clear that the superframe time duration depends not only on the cyclic prefix length but the system operating mode as well. This will affect the preamble design, as will be seen later. There is significantly more flexibility in MBWA as well as other emerging wireless technologies (e.g., 3GPP LTE) compared to existing 3G systems. Flexible parameters that can affect preamble structure are: •Bandwidth allocation which corresponds to a PHY frame FFT size of NFFT = 512/1024/2048 and the number of guard tones. This is a highly desirable feature that allows the operator to deploy the network in frequency bands with various sizes of available bandwidths. •FDD/TDD modes: FDD includes full and half duplex while TDD includes choice of TDD partitioning. This feature allows the operator to deploy the system in either FDD bands (paired) or TDD bands. FDD half duplex mode allows the system to serve lower-end terminals that only support half duplex communication. Configurable TDD partitioning allows the operator to fine tune the network performance according to the uplink and downlink traffic loadings. •Cyclic prefix length (four possible values): This allows the operator to choose a cyclic prefix length that is best suited to the expected delay spreads in the deployment. For example, in an urban area where a short delay spread is expected, a short CP can be selected, whereas in a suburb or a rural area where a long delay spread is typical, a long CP may then be used. •Synchronous/asynchronous modes: MBWA systems can operate in synchronous mode, where different sectors have access to a common timing reference such as the Global Positioning System (GPS), and asynchronous mode, where they do not. For example, for base stations mounted on a high-rise tower, synchronous mode can be selected to achieve the best overall network performance. For base stations situated inside a building where synchronization to the network is not possible, asynchronous mode may be activated. The flexibility in MBWA system configuration requires that the preamble be structured to provide an efficient means for system acquisition and determination for an access terminal. The widely variable bandwidths used in MBWA wireless systems, as well as the wide choices of deployment parameters, present significant challenges in acquisition system design. This article describes these challenges, as well as the solutions provided by the MBWA standard. The design scheme applies to 3GPP LTE and 3GPP2 UMB as well [4].
SB
IEEE
PBCCH
Communications
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
165
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Variable
Since acquisition
Fixed (max)
A
BEMaGS F
Variable
both MBWA and LTE have selected the narrowband acquisi-
PBCCH
a practical system,
Receive power
more importance for
PHY frame
H
CC
SB
Acquisition pilot 1 Acquisition pilot 2 Acquisition pilot 3
complexity is of
PHY frame
PHY frame
PHY frame
tion pilot design. The Time
preamble bandwidth of LTE is limited to
Figure 2. Illustration of terminal receiving power at various portions of the superframe.
1.28 MHz (the minimum bandwidth supported by LTE) independent of the system bandwidth.
166
Communications IEEE
central 5MHz of the system bandwidth, even when the system bandwidth of the deployment is more than 5 MHz, as is illustrated in Fig. 3. This design has several advantages. First and foremost, it significantly simplifies the acquisition complexity as the system bandwidth or FFT size, N FFT, may not be known a priori to the access terminal (AT) during initial acquisition. Second, it prevents “energy dilution” of time-domain signal taps. Since there are more distinguishable channel taps for a given channel in wider bandwidths, each such tap has lower energy when compared to a channel tap in a narrowband signal. During the acquisition of Acquisition Pilot 1, multi-paths are typically not combined for complexity reasons. This phenomenon, which we refer to as “energy dilution” can degrade the performance of any algorithm that attempts to look for channel taps in the time-domain. Restricting the Acquisition pilots to 5 MHz mitigates the effect of energy dilution. Third, it lowers complexity by allowing for correlations with shorter sequences (512 length sequences, as opposed to, for example, 2048 length sequences in 20 MHz). Finally, it helps in a faster initial acquisition, as an access terminal roaming between deployments of at least 5 MHz can always tune to the central 5 MHz and perform acquisition. The disadvantage of this design is the loss of wideband frequency diversity as opposed to a wideband preamble design. For this reason, in some configurations, the narrowband preamble is allowed to hop among the whole frequency band at the cost of increased receiver complexity. Another disadvantage is the loss of processing gain since wider bandwidth acquisition pilots have larger processing gain and thus can be detected further away from the transmitting base station. This is useful in mobile positioning as well as early discovery of neighbor cells. Since acquisition complexity is of more importance for a practical system, both MBWA and LTE have selected the narrowband acquisition pilot design. The preamble bandwidth of LTE is limited to 1.28 MHz (the minimum bandwidth supported by LTE) independent of the system bandwidth. The minimum bandwidth supported by MBWA is 2.5 MHz. The natural choice for the preamble bandwidth seems to be 2.5 MHz. However, MBWA optimized the design to 5 MHz bandwidth, believing that most next-generation system deployments have at least 5 MHz band-
width. The MBWA preamble thus has better cell penetration/coverage than LTE as a result of larger processing gain due to the wider preamble bandwidth.
ACQUISITION PILOT 1 Acquisition Pilot 1 is used for initial coarse timing (the position of the Acquisition Pilot 1 symbol) acquisition and frequency offset (between the access terminal and the access network) estimation. Since it is the very first pilot that a terminal searches for, its waveform • Should be made as simple as possible (i.e., less unknown parameters) to reduce the searching complexity. • Should tolerate frequency offset for detection since the access terminal’s frequency may not be synchronized to the access network at the beginning of the acquisition process. Acquisition Pilot 1 is transmitted on the 5th OFDM symbol in the preamble, spans the central subcarriers, and occupies every fourth subcarrier over this span, resulting in four copies of the same waveform in time domain. The use of four replicas of the same waveform instead of just one long waveform is to reduce the waveform period such that the access terminal’s frequency offset (relative to the access network) has less effect on the correlation performed by the access terminal during the detection for the Acquisition Pilot 1 waveform. In addition, having four periods is helpful for frequency offset estimation [9]. Acquisition Pilot 1 uses a frequency domain complex sequence, i.e., the generalized chirp-like (GCL) polyphase sequence [10] to modulate the subcarriers. One important property of a GCL sequence is that its Fourier transform also has a constant magnitude. Hence, the corresponding time domain waveform of each period has a constant magnitude that helps improve peak to average power ratio (PAPR). Low PAPR waveforms allow for a higher power amplifier setting at the transmitter, thereby extending coverage. It should be noted here that the coverage requirements for acquisition are typically higher than that for data traffic, since a mobile terminal should be able to acquire a sector before it is in the data coverage of a sector, thereby allowing for seamless handoff to that sector if required. Even if the access terminal detects Acquisition Pilot 1 (i.e., the terminal has located the
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Acquisition Pilot 1), the terminal still does not have the OFDM symbol boundary since the cyclic prefix length still remains unknown. Consequently, the terminal cannot determine the boundary of the following OFDM symbol (Acquisition Pilot 2). Therefore, Acquisition Pilot 1 should carry CP length information. This can be done by using different GCL sequences to indicate various cyclic prefix lengths. Since MBWA allows the use of four different cyclic prefix lengths (6.51, 13.02, 19.53, and 26.04 ms), four different waveforms are needed to indicate the four cyclic prefix values. Therefore, a terminal needs to perform four hypothesis tests to determine the cyclic prefix value used in the system. Information on cyclic prefix is necessary for obtaining symbol boundary, which is needed for detecting Acquisition Pilot 2. An alternative process used by LTE for generating time domain waveforms with constant profile is to directly use a time domain complex pseudo-noise (PN) sequence (a set of bits that are generated to be statistically random — see [11] for a detailed description of PN sequence), or a Zadoff-Chu complex sequence (a special case of the GCL sequence used in LTE), modulated QPSK waveform of length 128 and is repeated four times to form an Acquisition Pilot 1. For four values of cyclic prefix lengths, a PN sequence with four offset values can be used. Each offset represents a specific cyclic prefix value. A terminal needs to perform four hypothesis tests on four different PN offsets during the detection of Acquisition Pilot 1. Since inter-carrier frequency interval is fixed, once cyclic prefix length is determined, the complete OFDM symbol duration (including cyclic prefix) is known as well as the PHY frame duration. An alternative approach to embedding cyclic prefix information in Acquisition Pilot 1 is to leave the cyclic prefix hypothesis tests to the detection of Acquisition Pilot 2, i.e., Acquisition Pilot 1 does not include cyclic prefix information (one common waveform). After the detection of Acquisition Pilot 1, a terminal makes multiple hypothesis tests on the CP length to determine the starting position of Acquisition Pilot 2 when detecting Acquisition Pilot 2. LTE takes this approach. However, only three cyclic prefix lengths (4.7, 16.7 and 53.3 +s) are supported. This is a better design as compared to the MBWA approach, since there is only one waveform that a terminal needs to search for. The CP length is determined after Acquisition Pilot 1 is detected. To reduce the detection complexity, the same complex PN sequence of length 512 is used for the system with bandwidth less than 5 MHz (i.e., the number of usable subcarriers is less than 512). In this case, the complex PN sequence is FFT transformed to frequency domain and is used to QPSK modulate the 512 subcarriers. The unusable subcarriers or the guard subcarriers, are punctured. The effect of the guard subcarrier on PN sequence correlation properties will be discussed in the next section. The resulting frequency domain sequence is IFFT transformed back to time domain. Acquisition Plot 1 provides Acquisition Pilot 1 position and OFDM symbol duration (there-
Bandwidth < 5 MHz
5 MHz
IEEE
BEMaGS F
Bandwidth > 5 MHz
Figure 3. Illustration of narrowband preamble for various system bandwidths.
fore PHY frame length) information. Acquisition Pilot 1 waveform is common to all sectors that does not identify the sectors.
ACQUISITION PILOTS 2 AND 3 Acquisition Pilot 2 is a time domain Walsh sequence/code that carries the sector’s unique identification, i.e., the Pilot PN offset, which helps the access terminal to distinguish multiple sectors in the network. Walsh sequences are orthogonal binary sequences [11]. The advantage of Walsh-coding the Pilot PN offset is that the detection of a Walsh code can be efficiently implemented using fast Hadamard transforms (FHT), as will be seen later. The Pilot PN offset is a 9-bit identifier of a sector. In MBWA and LTE, a total of 29 = 512 PN offsets are used for sector identification. The Walsh sequence length is equal to the preamble FFT size with its index equal to the 9-bit Sector Pilot PN offset of value P between 0 and 511 (29 – 1). In addition, another complex PN sequence of length 512 (common to all sectors) is used to scramble the Walsh sequence. The resulting sequence is transformed to frequency domain via a 512-point FFT and is used to modulate all 512 subcarriers except the guard subcarriers (if the system bandwidth is less than 5 MHz). The need for the use of a PN sequence to scramble the Walsh sequence is due to the fact that Walsh sequences with different indices possess very different spectral properties (non white), and the insertion of guard subcarriers may destroy the Walsh code property depending on the Walsh code’s spectral property. The use of complex PN scrambling of Walsh sequence spreads the code energy evenly throughout the spectrum and, therefore, the insertion of guard carriers has more uniform effect on all Walsh sequences regardless of the individual Walsh code’s spectral property. Also note that the insertion of guard subcarriers has two effects on the time domain waveform. First, the constant modulus property is distorted; second, the correlation (cross and auto) properties of complex PN scrambled Walsh sequences are also impaired, as shown in Fig. 4. It is seen that both the cross and auto correlation properties of the Walsh sequence degrade as more guard carriers are inserted.
IEEE Communications Magazine • July 2011
Communications
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Preamble (max 5MHz)
IEEE
Frequency
Communications
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
167
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
1.0 0% 25% 50%
0.8
75%
CDF
0.6
0.4 Cross
Auto
0.2
0.0 0.0
0.2
0.4
0.6 0.8 Correlation magnitude
1.0
1.2
1.4
Figure 4. Change of cross/auto correlations CDFs of the PN scrambled Walsh sequence as a result of bandwidth reduction. The figure shows 0 reduction, 25 percent, 50 percent, and 75 percent reduction in bandwidth.
One may use different PN sequences, instead of Walsh sequences, to represent different sector Pilot PN offsets. However, the detection of Walsh sequence can be more efficient by applying the fast Hadamard transform (FHT), as opposed to performing PN correlation even with the help of FFT. This saving can be significant for the terminal receiver, especially when the number of selections is large (512 in this case). Like Acquisition Pilot 2, Acquisition Pilot 3 is also a time domain Walsh sequence, except that the Walsh sequence is scrambled by the sectorspecific complex PN sequence since the sector PN offset is available for the terminal after the detection of Acquisition Pilot 2. The 9-bit Walsh sequence index contains 9-bit system information and includes synchronous/asynchronous mode (1 bit), four least significant bits (LSBs) of the current superframe index, FDD/TDD mode (1 bit), full/half duplex mode (1 bit), TDD partitioning (1 bit), preamble frequency hopping mode, etc, that is necessary for decoding the PBCCH packet.
PBCCH AND SBCCH After the Acquisition Pilots are detected, system frequency and coarse timing are established. OFDM signaling and channel coding can then be used to convey larger amounts of system information more efficiently. The PBCCH symbol contains the information packet that is channel coded and OFDM modulated. However, since the bandwidth information is not yet known to the access terminal, the PBCCH packet is coded with very low effective rate. The PBCCH packet contains the system information, including the complete superframe index (system time) at the time the PBCCH packet is generated and deployment-wide static parameters such as total number of subcarriers, number of guard subcarriers (in units of 16), etc. Each PBCCH packet is
168
Communications IEEE
A
BEMaGS F
CRC (12 bits) protected, channel encoded (1/3 code rate), interleaved, repeated, scrambled with the sector Pilot PN, and QPSK modulated. The QPSK symbols are then mapped to OFDM subcarriers. One natural way for designing the QPSK symbol to subcarrier mapping is to map the symbols only to usable subcarriers, resulting in a mapping that is bandwidth dependent. A better design is to map the QPSK symbols to 512 subcarriers even if some of the subcarriers may not be usable (guard carriers). The guard subcarriers (if any) are then punctured (zeroed out). This way, the mapping of the modulation symbols to the subcarriers is independent from the actual bandwidth of the preamble or the number of guard subcarriers. The PBCCH modulation symbols can thus be de-mapped without knowing the number of guard subcarriers, which allows the PBCCH packets to be decoded without the full knowledge of the bandwidth. The static nature of the PBCCH packet allows the transmission of the PBCCH packet with low effective coding rate without high overhead. This is achieved by updating the content of the PBCCH packet every 16 superframes. The same PBCCH packet is repeatedly transmitted over 16 consecutive superframes (four repetitions in LTE). That is, the PBCCH packet is updated when the superframe index modulo 16 is equal to zero, or, the four LSBs of the superframe index is equal to zero. This allows efficient incremental redundancy decoding at the terminal, as will be seen in the next section. The Secondary Broadcast Channel (SBCCH) is carried on the OFDM symbols with indices 1 through 4 of the preamble in a superframe. The SBCCH channel is designed under the assumption that the PBCCH packet has been successfully decoded. The system bandwidth (and therefore the preamble bandwidth) is known. A more efficient coding (higher code rate) can then be used for SBCCH, allowing an SBCCH packet to be much larger than the PBCCH packet. A SBCCH packet contains further detailed system configuration information that a terminal needs for accessing the network, such as the number of effective antennas, common pilot channel hopping mode, common pilot spacing, and number of sub-trees for SDMA, etc. It is appended with CRC (12 bits), channel encoded, interleaved, repeated, scrambled with sector PN, and QPSK modulated onto usable subcarriers. The SBCCH packet is updated every superframe.
PREAMBLE ACQUISITION The goal of the preamble acquisition is to acquire the system synchronization information and the system parameters necessary to access the system from the preamble at a given carrier frequency. The IEEE MBWA system acquisition procedure is summarized in Fig. 5.
ACQUISITION PILOT 1 DETECTION The MBWA system acquisition begins with searching for Acquisition Pilot 1. At a given carrier frequency, the access terminal looks for the Acquisition Pilot 1 signal by correlating the received signal with each of the four hypotheses
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
A
BEMaGS
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Start
Detect acquisition pilot 1
Detect acquisition pilot 2
Detect acquisition pilot 3
Decode PBCCH
No Decode success? Yes
Decode SBCCH
that non-coherent combining of four short correlation periods instead of a single long block correlation is due to the potential high frequency offset between the access terminal and the network in the initial system acquisition. A frequency offset results in phase rotation or shift to the sampled signals, causing reduced correlation performance between the received signal and the hypothesis waveform. The longer the correlation period, the larger the phase shift and the poorer the correlation performance. A shorter correlation period thus helps mitigate the frequency offset effect. However, for the non-initial acquisition case (e.g., neighbor cell search) where access terminals are synchronized to the network, performing one single block correlation for the whole OFDM symbol duration is actually preferred for greater processing gain. What does a terminal gain from the detection of Acquisition Pilot 1? First, the terminal’s frequency offset can be estimated from the four periods of the Acquisition Pilot 1 and corrected from the terminal’s local frequency synthesizer. A frequency offset estimation method using four signal segments is described in [9]. Second, the access terminal gains the knowledge of the cyclic prefix duration. Therefore, the OFDM symbol boundary is determined. The Acquisition Pilot 2 can now be located as a result. A coarse timing and frequency synchronization to the system has now been established. In addition, once the Acquisition Pilot 1 is detected, one can also search for multi-paths within a certain range that Acquisition Pilot 1 is detected for use in Acquisition Pilots 2 and 3 for multi-path diversity gain.
F
For the non-initial acquisition case (e.g., neighbor cell search) where access terminals are synchronized to the network, performing one single block correlation for the whole OFDM symbol duration is actually preferred for greater processing gain.
ACQUISITION PILOT 2 PROCESSING End
Figure 5. Flowchart of IEEE MBWA system acquisition procedure.
over the duration of at least one superframe until one candidate is detected. That is, a moving sum of one of the four hypothesis correlations corresponding to the four periods of Acquisition Pilot 1 waveform is performed and compared to a threshold. Without knowing the actual preamble bandwidth, the received signal is first sampled at the full preamble bandwidth (i.e., 5MHz). Since correlation can be done more efficiently with FFT, a 128-point FFT is applied to each of the four period samples, and spectrum-shaped assuming minimum bandwidth (i.e., zero out the maximum possible guard subcarriers).1 The resulting frequency domain samples are then multiplied with one of the hypothesis Acquisition Pilot 1 waveforms in the frequency domain to get the correlation values. The four correlation values are non-coherently combined across four consecutive FFT windows and compared with a threshold to determine if an Acquisition Pilot 1 is present. The minimum bandwidth of 2.5 MHz assumption results in a maximum 3dB detection performance loss. Note
With the correction of the frequency offset and the knowledge of Acquisition Pilot 2 location, the whole Acquisition Pilot 2 symbol can be sampled at once at the full preamble bandwidth. The sampled data are first transformed to frequency domain via a 512-point FFT. As with the Acquisition Pilot 1, the frequency domain data are spectrum-shaped. The resulting data are then transformed back to the time domain sequence, descrambled, and a fast Hadamard transform (FHT) is used on the descrambled data to detect the Walsh sequence for the sector Pilot PN offset. In detail, the paths crossing the threshold obtained from Acquisition Pilot 1 processing over one superframe are passed on for Acquisition Pilot 2 processing, which involves performing the FHT and comparing the resulting sector energies to a threshold. The FHT threshold is chosen by design to maintain a false alarm probability within a desirable level. A false alarm event is defined as a threshold-crossing occurring in an empty channel, i.e. noise-only scenario. A false alarm event would incur a penalty time. This penalty is attributed to unnecessary attempts at decoding system information following a false alarm event. The FHT effectively performs correlations with each of the Walsh sequences and the output of the FHT corresponding to the Walsh code with an index value equal to the sector’s 9-bit PN offset.
1
Note that for certain deployment with minimum bandwidth larger than 5 MHz, spectrum shaping is then not needed.
IEEE Communications Magazine • July 2011
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
169
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Upon the detection of Acquisition Pilot 2, the access terminal obtains the sector’s Pilot PN offset.
ACQUISITION PILOT 3 PROCESSING Acquisition Pilot 3 is next sampled at the 5 MHz bandwidth, spectrum-shaped and descrambled using the Pilot PN detected from Acquisition Pilot 2. Like the processing of Acquisition Pilot 2, the FHT is then applied to the descrambled data to detect the acquisition information. Upon the detection of Acquisition Pilot 3, the acquisition information including synchronous/ asynchronous mode, 4 LSBs of the superframe index, full/half duplex modes, FDD/TDD mode, and TDD partitioning (if TDD mode), is avail-
A
BEMaGS F
able to the access terminal. Based on this information, the access terminal is now able to determine the superframe boundary. Together with the additional knowledge of AGC gain, the terminal is now able to locate and acquire the PBCCH and SBCCH symbols in the next superframe preamble. The overall detection performance of the Acquisition Pilots is given in Fig. 6, where geometry is defined as the ratio of signal power over interference. Geometry of 0 dB thus indicates the cell boundary. It is seen that the Acquisition Pilots can be detected at very low geometry, i.e., deep penetration into neighbor cell. This is a desirable feature to facilitate early discovery of new neighbor cells for potential hand off, etc.
PBCCH DECODING 16 Pedestrian B (3 km/h) Vehicular A (120 km/h) Detection time (number of preambles)
14 12 10 8 6 4 2 0 -14 -13 -12 -11 -10
-9
-8 -7 -6 -5 Geometry (dB)
-4
-3
-2
-1
0
Figure 6. Acquisition Pilots 1, 2, 3 joint detection performance (95 percentile, joint false alarm probability = 0.001, 1 receive antenna, 5MHz bandwidth at 2 GHz carrier frequency, 10 ppm receiver frequency offset).
1t sm
ons
on
issi
issi
sm
ns
ns
issio
issio
nsm
nsm
ran
ran
2t
4 tra
0.1
8 tra
PER
1
0.01 -15
-13
-11
-9
-7 -5 Geometry (dB)
-3
-1
Figure 7. PBCCH decoding performance at various levels of redundancies (channel model: Pedestrian Type B 3km/h, one receive antenna).
170
Communications IEEE
1
After the detection of the Acquisition Pilots, the access terminal is ready to decode the PBCCH packet contained in the next superframe preamble. With the acquired superframe timing, the access terminal is able to locate the PBCCH OFDM symbol in the following superframe preamble. The AGC used to acquire Acquisition Pilots is set for sampling the PBCCH OFDM symbol. Note that the coarse timing acquired from the Acquisition Pilot 1 is good for time domain waveform detection (e.g., Acquisition Pilots 2 and 3 detection) but is not sufficient for the PBCCH OFDM symbol demodulation. Before the PBCCH OFDM symbol can be sampled, the optimal OFDM symbol timing (i.e., the optimal FFT collection window) should be established to minimize the multi-path effect using the OFDM symbol timing technique described in [12]. The PBCCH OFDM symbol is then sampled at the full preamble bandwidth and performs FFT with the preamble FFT size of 512. The frequency domain data are spectrum-shaped, de-mapped, demodulated, descrambled, de-interleaved, log likelihood ratio (LLR) calculated and decoded. As in Acquisition Pilot 1, 2, and 3 detection, the conservative spectrum-shaping may result in loss of SINR up to 3 dB. However, as will be seen later, a PBCCH packet is coded with very low code rate. The loss of 3 dB does not prevent PBCCH from being successfully decoded. A failure to decode a PBCCH packet is most likely due to insufficient SINR. Therefore, if the decoding is not successful, the access terminal determines if the PBCCH carries the last transmission of the 16 transmissions by checking if the four LSBs of the superframe index is equal to 15. If the current received PBCCH is not the last of the 16 transmissions, the LLR from the successive transmission of the PBCCH are combined with the LLR stored in the LLR buffer and another decoding attempt is made. Otherwise, the buffer is cleared and the LLR data are discarded. This procedure is repeated until a successful decoding. The maximum number of transmissions the access terminal can combine is 16 since the PBCCH packet is updated every 16 superframes. As shown in Fig. 7, the decoding of a PBCCH packet rarely takes all 16 transmissions. High geometry users are more likely to need less redundancy for less processing gain to decode the packet as compared to edge users. It there-
IEEE Communications Magazine • July 2011
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
fore takes less time for high geometry users to acquire the system thereby reducing the acquisition time. A decoding failure may also be the consequence of a false detection of Acquisition Pilots 1 to 3. In the case that decoding fails, even after the LLR buffer has combined 16 consecutive PBCCH transmissions, the acquisition procedure restarts. Upon a successful decoding of the PBCCH packet, the access terminal confirms that the information acquired from Acquisition Pilots 1 to 3 is correct and a MBMA system indeed exists at this carrier frequency. In addition, it obtains the system information including superframe index, system FFT size, and number of guard subcarriers, etc. from the PBCCH packet. This information is necessary for decoding the following SBCCH packet.
SBCCH DECODING Once PBCCH is decoded, the access terminal starts to acquire SBCCH. The FFT collection window for the four SBCCH OFDM symbols is determined. Four OFDM symbols from 1 to 4 are then sampled and transformed to frequency domain via an FFT. Using the number of guard subcarrier information from the PBCCH, the actual guard subcarriers are zeroed out, the modulation symbols are demodulated, descrambled, de-interleaved and decoded. By now the access terminal has all the information necessary to access the system (e.g., sending an access probe on the reverse link access channel) and the preamble acquisition process is completed. The access terminal can now notify its existence to the network by performing access probing through the uplink random access channel (RACH).
System determination is a set of data structures and protocols to serve the purpose of identifying the best system that is suitable for a terminal to operate in a given environment. It consists of the physical layer system acquisition (described in the previous section) and the upper layer system selection. System determination is performed during: • Terminal power up. • Terminal rescan for better service. • Out of service. During system determination, a terminal has to perform acquisition scan, i.e., to perform acquisition at all potential carrier frequencies with channel raster of 200 kHz. Channel raster of 200 kHz means that the carrier center frequency must be at an integer multiple of 200 kHz as specified by the standard. It is clear that performing a whole band acquisition scan consumes a lot of time and power. To accelerate the system determination process, a Preferred Roaming List (PRL) is used, which is an operator supplied list of systems that defines the systems that a mobile can access. The PRL is programmed into a terminal’s nonvolatile device prior to distribution, or at an operator’s service center, or via an over the air protocol.
Index
Type
Band
0
WBMA
PCS 75
1
WCDMA
AWS 21, GSM 850, 1900
2
PCS CDMA
PCS 100, 125,150,175
3
Cellular CDMA
System A
4
Cellular Analog
System B
IEEE
F
System Table SID
GEO1
Priority2
Acquisition Index
111
0
1
0
34
1
0
3
400
0
1
1
4
1
1
0
12
1
1
3
0
1
0
4
61
0
1
0
56
1
1
1
16
1
0
4
1
“0” indicates the start of a new GEO. “1” has higher priority than “0”.
Table 1. A sample preferred roaming list (PRL).
A PRL consists of an Acquisition Table and a System Table. An Acquisition Table contains an indexed (ordered according to preference) list of RF channels to search. Each entry/record describes the RF environment. A system table contains a list of system descriptions keyed by system identification (SID). Each entry/record is part of a geographic area (GEO). Preference can exist within geographic areas. An operator can specify preferences for which networks to access (Table 1). PRL assists a terminal with system acquisition and selection by providing information on the relevant radio access technologies and how to acquire them efficiently. Upon system acquisition, the terminal uses PRL to determine whether the acquired system is usable or not, whether the system is the optimal system on which to operate in the given location, and what are the systems that are more preferred and how to acquire them efficiently if the current system is not optimal/most preferred. As shown in Fig. 8, upon powering up, the access terminal builds a list of channels to perform the acquisition scan, i.e., the terminal builds a list of systems consisting of, first, all the
IEEE Communications Magazine • July 2011
Communications
BEMaGS
Acquisition Table
2
SYSTEM DETERMINATION
A
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
171
A
BEMaGS F
Communications IEEE
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
Power-up
Build acquisition list
MRU/PRL
Acquisition scan
Out of service search
N
Service found?
Frequency scan
N
Service found?
Register to the system MRU
Background search
Update MRU
N
Most preferred system?
Exit
CONCLUSION
172
Communications IEEE
F
APPENDIX In this appendix, a mathematical analysis of the detection of the Acquisition Pilot signals is given. Since Acquisition Pilots 2 and 3 have the same structure, the analysis of Acquisition Pilot 3 is thus omitted.
SYSTEM MODEL Transmitted Signal: Acquisition Pilot 1, 2 — The transmitted GCL sequence in the OFDM symbol corresponding to Acquisition Pilot 1 in the superframe preamble occupies every fourth subcarrier. More precisely, it is given by £ k ( k + 1) ¥ N P4 k = exp ² < j 2/ u , 0 ) k < FFT ´ 2 NG ¦ 4 ¤
(1)
where N FFT = 512 is the preamble number of subcarriers, NG = NFFT/4 – 1 = 127, and parameter u is a function of NFFT and the cyclic prefix length [4]. To simplify the analysis, we assume the number of guard carriers is zero. It can be shown that the corresponding time domain waveform of each period is £ £ k ( k + 1) ¥ n < n 2 / u ¥ NG