Network Working Group                                           S. Floyd
Request for Comments: 3742                                          ICSI
Category: Experimental                                        March 2004


       Limited Slow-Start for TCP with Large Congestion Windows

Status of this Memo

  This memo defines an Experimental Protocol for the Internet
  community.  It does not specify an Internet standard of any kind.
  Discussion and suggestions for improvement are requested.
  Distribution of this memo is unlimited.

Copyright Notice

  Copyright (C) The Internet Society (2004).  All Rights Reserved.

Abstract

  This document describes an optional modification for TCP's slow-start
  for use with TCP connections with large congestion windows.  For TCP
  connections that are able to use congestion windows of thousands (or
  tens of thousands) of MSS-sized segments (for MSS the sender's
  MAXIMUM SEGMENT SIZE), the current slow-start procedure can result in
  increasing the congestion window by thousands of segments in a single
  round-trip time.  Such an increase can easily result in thousands of
  packets being dropped in one round-trip time.  This is often
  counter-productive for the TCP flow itself, and is also hard on the
  rest of the traffic sharing the congested link.  This note describes
  Limited Slow-Start as an optional mechanism for limiting the number
  of segments by which the congestion window is increased for one
  window of data during slow-start, in order to improve performance for
  TCP connections with large congestion windows.

1.  Introduction

  This note describes an optional modification for TCP's slow-start for
  use with TCP connections with large congestion windows.  For TCP
  connections that are able to use congestion windows of thousands (or
  tens of thousands) of MSS-sized segments (for MSS the sender's
  MAXIMUM SEGMENT SIZE), the current slow-start procedure can result in
  increasing the congestion window by thousands of segments in a single
  round-trip time.  Such an increase can easily result in thousands of
  packets being dropped in one round-trip time.  This is often
  counter-productive for the TCP flow itself, and is also hard on the
  rest of the traffic sharing the congested link.  This note describes
  Limited Slow-Start, limiting the number of segments by which the



Floyd                         Experimental                      [Page 1]

RFC 3742     TCP's Slow-Start with Large Congestion Windows   March 2004


  congestion window is increased for one window of data during slow-
  start, in order to improve performance for TCP connections with large
  congestion windows.

  When slow-start results in a large increase in the congestion window
  in one round-trip time, a large number of packets might be dropped in
  the network (even with carefully-tuned active queue management
  mechanisms in the routers).  This drop of a large number of packets
  in the network can result in unnecessary retransmit timeouts for the
  TCP connection.  The TCP connection could end up in the congestion
  avoidance phase with a very small congestion window, and could take a
  large number of round-trip times to recover its old congestion
  window.  This poor performance is illustrated in [F02].

2.  The Proposal for Limited Slow-Start

  Limited Slow-Start introduces a parameter, "max_ssthresh", and
  modifies the slow-start mechanism for values of the congestion window
  where "cwnd" is greater than "max_ssthresh".  That is, during Slow-
  Start, when

     cwnd <= max_ssthresh,

  cwnd is increased by one MSS (MAXIMUM SEGMENT SIZE) for every
  arriving ACK (acknowledgement) during slow-start, as is always the
  case.  During Limited Slow-Start, when

     max_ssthresh < cwnd <= ssthresh,

  the invariant is maintained so that the congestion window is
  increased during slow-start by at most max_ssthresh/2 MSS per round-
  trip time.  This is done as follows:

     For each arriving ACK in slow-start:
       If (cwnd <= max_ssthresh)
          cwnd += MSS;
       else
          K = int(cwnd/(0.5 max_ssthresh));
          cwnd += int(MSS/K);

  Thus, during Limited Slow-Start the window is increased by 1/K MSS
  for each arriving ACK, for K = int(cwnd/(0.5 max_ssthresh)), instead
  of by 1 MSS as in standard slow-start [RFC2581].








Floyd                         Experimental                      [Page 2]

RFC 3742     TCP's Slow-Start with Large Congestion Windows   March 2004


  When

    ssthresh < cwnd,

  slow-start is exited, and the sender is in the Congestion Avoidance
  phase.

  Our recommendation would be for max_ssthresh to be set to 100 MSS.
  (This is illustrated in the NS [NS] simulator, for snapshots after
  May 1, 2002, in the tests "./test-all-tcpHighspeed tcp1A" and
  "./test-all-tcpHighspeed tcpHighspeed1" in the subdirectory
  "tcl/lib".  Setting max_ssthresh to Infinity causes the TCP
  connection in NS not to use Limited Slow-Start.)

  With Limited Slow-Start, when the congestion window is greater than
  max_ssthresh, the window is increased by at most 1/2 MSS for each
  arriving ACK; when the congestion window is greater than 1.5
  max_ssthresh, the window is increased by at most 1/3 MSS for each
  arriving ACK, and so on.

  With Limited Slow-Start it takes:

     log(max_ssthresh)

  round-trip times to reach a congestion window of max_ssthresh, and it
  takes:

     log(max_ssthresh) + (cwnd - max_ssthresh)/(max_ssthresh/2)

  round-trip times to reach a congestion window of cwnd, for a
  congestion window greater than max_ssthresh.

  Thus, with Limited Slow-Start with max_ssthresh set to 100 MSS, it
  would take 836 round-trip times to reach a congestion window of
  83,000 packets, compared to 16 round-trip times without Limited
  Slow-Start (assuming no packet drops).  With Limited Slow-Start, the
  largest transient queue during slow-start would be 100 packets;
  without Limited Slow-Start, the transient queue during Slow-Start
  would reach more than 32,000 packets.

  By limiting the maximum increase in the congestion window in a
  round-trip time, Limited Slow-Start can reduce the number of drops
  during slow-start, and improve the performance of TCP connections
  with large congestion windows.







Floyd                         Experimental                      [Page 3]

RFC 3742     TCP's Slow-Start with Large Congestion Windows   March 2004


3.  Experimental Results

  Tom Dunigan has added Limited Slow-Start to the Linux 2.4.16 Web100
  kernel, and performed experiments comparing TCP with and without
  Limited Slow-Start [D02].  Results so far show improved performance
  for TCPs using Limited Slow-Start.  There are also several
  experiments comparing different values for max_ssthresh.

4.  Related Proposals

  There has been considerable research on mechanisms for the TCP sender
  to learn about the limitations of the available bandwidth, and to
  exit slow-start before receiving a congestion indication from the
  network [VEGAS,H96].  Other proposals set TCP's slow-start parameter
  ssthresh based on information from previous TCP connections to the
  same destination [WS95,G00].  This document proposes a simple
  limitation on slow-start that can be effective in some cases even in
  the absence of such mechanisms.  The max_ssthresh parameter does not
  replace ssthresh, but is an additional parameter.  Thus, Limited
  Slow-Start could be used in addition to mechanisms for setting
  ssthresh.

  Rate-based pacing has also been proposed to improve the performance
  of TCP during slow-start [VH97,AD98,KCRP99,ASA00].  We believe that
  rate-based pacing could be of significant benefit, and could be used
  in addition to the Limited Slow-Start in this proposal.

  Appropriate Byte Counting [RFC3465] proposes that TCP increase its
  congestion window as a function of the number of bytes acknowledged,
  rather than as a function of the number of ACKs received.
  Appropriate Byte Counting is largely orthogonal to this proposal for
  Limited Slow-Start.

  Limited Slow-Start is also orthogonal to other proposals to change
  mechanisms for exiting slow-start.  For example, FACK TCP includes an
  overdamping mechanism to decrease the congestion window somewhat more
  aggressively when a loss occurs during slow-start [MM96].  It is also
  true that larger values for the MSS would reduce the size of the
  congestion window in units of MSS needed to fill a given pipe, and
  therefore would reduce the size of the transient queue in units of
  MSS.

5.  Acknowledgements

  This proposal is part of a larger proposal for HighSpeed TCP for TCP
  connections with large congestion windows, and resulted from
  simulations done by Evandro de Souza, in joint work with Deb Agarwal.
  This proposal for Limited Slow-Start draws in part from discussions



Floyd                         Experimental                      [Page 4]

RFC 3742     TCP's Slow-Start with Large Congestion Windows   March 2004


  with Tom Kelly, who has used a similar modified slow-start in his own
  research with congestion control for high-bandwidth connections.  We
  also thank Tom Dunigan for his experiments with Limited Slow-Start.

  We thank Andrei Gurtov, Reiner Ludwig, members of the End-to-End
  Research Group, and members of the Transport Area Working Group, for
  feedback on this document.

6.  Security Considerations

  This proposal makes no changes to the underlying security of TCP.

7.  References

7.1.  Normative References

  [RFC2581] Allman, M., Paxson, V. and W. Stevens, "TCP Congestion
            Control", RFC 2581, April 1999.

  [RFC3465] Allman, M., "TCP Congestion Control with Appropriate Byte
            Counting (ABC)", RFC 3465, February 2003.

7.2.  Informative References

  [AD98]    Mohit Aron and Peter Druschel, "TCP: Improving Start-up
            Dynamics by Adaptive Timers and Congestion Control"",
            TR98-318, Rice University, 1998.  URL "http://cs-
            tr.cs.rice.edu/Dienst/UI/2.0/Describe/ncstrl.rice_cs/TR98-
            318/".

  [ASA00]   A. Aggarwal, S. Savage, and T. Anderson, "Understanding the
            Performance of TCP Pacing", Proceedings of the 2000 IEEE
            Infocom Conference, Tel-Aviv, Israel, March, 2000.  URL
            "http://www.cs.ucsd.edu/~savage/".

  [D02]     T. Dunigan, "Floyd's TCP slow-start and AIMD mods", 2002.
            URL "http://www.csm.ornl.gov/~dunigan/net100/floyd.html".

  [F02]     S. Floyd, "Performance Problems with TCP's Slow-Start",
            2002.  URL "http://www.icir.org/floyd/hstcp/slowstart/".

  [G00]     A. Gurtov, "TCP Performance in the Presence of Congestion
            and Corruption Losses", Master's Thesis, University of
            Helsinki, Department of Computer Science, Helsinki,
            December 2000.  URL
            "http://www.cs.helsinki.fi/u/gurtov/papers/ms_thesis.html".





Floyd                         Experimental                      [Page 5]

RFC 3742     TCP's Slow-Start with Large Congestion Windows   March 2004


  [H96]     J. C. Hoe, "Improving the Start-up Behavior of a Congestion
            Control Scheme for TCP", SIGCOMM 96, 1996.  URL
            "http://www.acm.org/sigcomm/sigcomm96/program.html".

  [KCRP99]  J. Kulik, R. Coulter, D. Rockwell, and C. Partridge, "A
            Simulation Study of Paced TCP", BBN Technical Memorandum
            No. 1218, 1999.  URL
            "http://www.ir.bbn.com/documents/techmemos/index.html".

  [MM96]    M. Mathis and J. Mahdavi, "Forward Acknowledgment: Refining
            TCP Congestion Control", SIGCOMM, August 1996.

  [NS]      The Network Simulator (NS). URL
            "http://www.isi.edu/nsnam/ns/".

  [VEGAS]   Vegas Web Page, University of Arizona.  URL
            "http://www.cs.arizona.edu/protocols/".

  [VH97]    Vikram Visweswaraiah and John Heidemann, "Rate Based Pacing
            for TCP", 1997.  URL
            "http://www.isi.edu/lsam/publications/rate_based_pacing/".

  [WS95]    G. Wright and W. Stevens, "TCP/IP Illustrated", Volume 2,
            Addison-Wesley Publishing Company, 1995.

Authors' Address

  Sally Floyd
  ICIR (ICSI Center for Internet Research)

  Phone: +1 (510) 666-2989
  EMail: [email protected]
  URL: http://www.icir.org/floyd/


















Floyd                         Experimental                      [Page 6]

RFC 3742     TCP's Slow-Start with Large Congestion Windows   March 2004


Full Copyright Statement

  Copyright (C) The Internet Society (2004).  This document is subject
  to the rights, licenses and restrictions contained in BCP 78 and
  except as set forth therein, the authors retain all their rights.

  This document and the information contained herein are provided on an
  "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE
  REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE
  INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR
  IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF
  THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
  WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Intellectual Property

  The IETF takes no position regarding the validity or scope of any
  Intellectual Property Rights or other rights that might be claimed
  to pertain to the implementation or use of the technology
  described in this document or the extent to which any license
  under such rights might or might not be available; nor does it
  represent that it has made any independent effort to identify any
  such rights.  Information on the procedures with respect to
  rights in RFC documents can be found in BCP 78 and BCP 79.

  Copies of IPR disclosures made to the IETF Secretariat and any
  assurances of licenses to be made available, or the result of an
  attempt made to obtain a general license or permission for the use
  of such proprietary rights by implementers or users of this
  specification can be obtained from the IETF on-line IPR repository
  at http://www.ietf.org/ipr.

  The IETF invites any interested party to bring to its attention
  any copyrights, patents or patent applications, or other
  proprietary rights that may cover technology that may be required
  to implement this standard.  Please address the information to the
  IETF at [email protected].

Acknowledgement

  Funding for the RFC Editor function is currently provided by the
  Internet Society.









Floyd                         Experimental                      [Page 7]


You are viewing proxied material from gopher.meulie.net. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.