Removing exponential backoff from TCP

The well-accepted wisdom is that TCP's exponential backoff mechanism, introduced by Jacobson 20 years ago, is essential for preserving the stability of the Internet. In this paper, we show that removing exponential backoff from TCP altogether can be done without inducing any stability side-effects. We introduce the implicit packet conservation principle and show that as long as the endpoints uphold this principle, they can only improve their end-to-end performance relative to the exponential backoff case. By conducting large-scale simulations, modeling, and network experiments in Emulab and the Internet using a kernel-level FreeBSD TCP implementation, realistic traffic distributions, and complex network topologies, we demonstrate that TCP's binary exponential backoff mechanism can be safely removed. Moreover, we show that insuitability of TCP's exponential backoff is fundamental, i.e., independent from the currently-dominant Internet traffic properties or bottleneck capacities. Surprisingly, our results indicate that a path to incrementally deploying the change does exist.

[1]  Mischa Schwartz,et al.  ACM SIGCOMM computer communication review , 2001, CCRV.

[2]  Sally Floyd,et al.  Promoting the use of end-to-end congestion control in the Internet , 1999, TNET.

[3]  Sampath Kannan,et al.  A Bound on the Capacity of Backoff and Acknowledgement-Based Protocols , 2000, ICALP.

[4]  Sally Floyd,et al.  Patterns of Congestion Collapse , 2003 .

[5]  Cheng Jin,et al.  FAST TCP: Motivation, Architecture, Algorithms, Performance , 2006, IEEE/ACM Transactions on Networking.

[6]  Robert Tappan Morris,et al.  TCP behavior with many flows , 1997, Proceedings 1997 International Conference on Network Protocols.

[7]  Robert Metcalfe,et al.  Ethernet: distributed packet switching for local computer networks , 1988, CACM.

[8]  John Nagle,et al.  Congestion control in IP/TCP internetworks , 1984, CCRV.

[9]  Frank Kelly,et al.  Stochastic Models of Computer Communication Systems , 1985 .

[10]  Michalis Faloutsos,et al.  Is P2P dying or just hiding? [P2P traffic measurement] , 2004, IEEE Global Telecommunications Conference, 2004. GLOBECOM '04..

[11]  Liang Guo,et al.  The war between mice and elephants , 2001, Proceedings Ninth International Conference on Network Protocols. ICNP 2001.

[12]  Walter Willinger,et al.  A first-principles approach to understanding the internet's router-level topology , 2004, SIGCOMM '04.

[13]  Priya Mahadevan,et al.  Orbis: rescaling degree correlations to generate annotated internet topologies , 2007, SIGCOMM 2007.

[14]  Priya Mahadevan,et al.  Orbis: rescaling degree correlations to generate annotated internet topologies , 2007, SIGCOMM '07.

[15]  David J. Aldous.  Ultimate instability of exponential back-off protocol for acknowledgment-based transmission control of random access communication channels , 1987, IEEE Trans. Inf. Theory.

[16]  Jin Cao,et al.  Stochastic models for generating synthetic HTTP source traffic , 2004, IEEE INFOCOM 2004.

[17]  Sally Floyd,et al.  Increasing TCP's Initial Window , 1998, RFC.

[18]  Abdesselem Kortebi,et al.  Evaluating the number of active flows in a scheduler realizing fair statistical bandwidth sharing , 2005, SIGMETRICS '05.

[19]  F. Donelson Smith,et al.  A Performance Study of Loss Detection/Recovery in Real-world TCP Implementations , 2007, 2007 IEEE International Conference on Network Protocols.

[20]  Michael Walfish,et al.  DDoS defense by offense , 2006, SIGCOMM 2006.

[21]  Sally Floyd,et al.  HighSpeed TCP for Large Congestion Windows , 2003, RFC.

[22]  Marcel Dischinger,et al.  Characterizing residential broadband networks , 2007, IMC '07.

[23]  Craig Partridge,et al.  Improving round-trip time estimates in reliable transport protocols , 1991, TOCS.

[24]  Kevin Jeffay,et al.  The effects of active queue management on web performance , 2003, SIGCOMM '03.

[25]  Aleksandar Kuzmanovic,et al.  The power of explicit congestion notification , 2005, SIGCOMM '05.

[26]  Alex C. Snoeren,et al.  Decongestion Control , 2006, HotNets.

[27]  Thomas E. Anderson,et al.  PCP: Efficient Endpoint Congestion Control , 2006, NSDI.

[28]  Vern Paxson,et al.  Computing TCP's Retransmission Timer , 2000, RFC.

[29]  Vern Paxson,et al.  On estimating end-to-end network path properties , 2001, SIGCOMM LA '01.

[30]  Sampath Kannan,et al.  A bound on the capacity of backoff and acknowledgment-based protocols , 2004, SIAM J. Comput..

[31]  Sally Floyd,et al.  Quick-Start for TCP and IP , 2007, RFC.

[32]  Anees Shaikh,et al.  An empirical evaluation of wide-area internet bottlenecks , 2003, SIGMETRICS '03.

[33]  Markku Kojo,et al.  Forward RTO-Recovery (F-RTO): An Algorithm for Detecting Spurious Retransmission Timeouts with TCP and the Stream Control Transmission Protocol (SCTP) , 2005, RFC.

[34]  Aleksandar Kuzmanovic,et al.  Pong: diagnosing spatio-temporal internet congestion properties , 2007, SIGMETRICS '07.

[35]  Limin Wang,et al.  Congestion Control Without a Startup Phase , 2010 .

[36]  Jia Wang,et al.  Locating internet bottlenecks: algorithms, measurements, and implications , 2004, SIGCOMM '04.

[37]  Tom Kelly,et al.  Scalable TCP: improving performance in highspeed wide area networks , 2003, CCRV.

[38]  Nick McKeown,et al.  Why flow-completion time is the right metric for congestion control , 2006, CCRV.

[39]  Guido Appenzeller,et al.  Sizing router buffers , 2004, SIGCOMM '04.

[40]  Scott Shenker,et al.  Endpoint admission control: architectural issues and performance , 2000, SIGCOMM.

[41]  Sally Floyd,et al.  Measuring the evolution of transport protocols in the internet , 2005, CCRV.

[42]  Yin Zhang,et al.  On individual and aggregate TCP performance , 1999, Proceedings. Seventh International Conference on Network Protocols.

[43]  Van Jacobson,et al.  Congestion avoidance and control , 1988, SIGCOMM '88.