Internet Engineering Task Force                                 S. Floyd
INTERNET-DRAFT                                                      ICIR
draft-floyd-tmrg-rtt-fairness-00a.txt                   22 November 2006
Expires: May 2007


               Metrics for Congestion Control Mechanisms:
    Why Round-Trip-Time Fairness would be Bad for Congestion Control


Abstract

    In TCP, bandwidth usage between competing flows is a function of the
    round-trip time, with short flows receiving more bandwidth than
    longer ones [F02] (Section 3.3).  RTT fairness refers to a goal of
    equal bandwidth usage between competing flows with different round-
    trip times, for competing flows that share the same congested links.
    As an example, one way to achieve RTT fairness between TCP
    connections would be for each TCP sender to use a Constant-Rate
    window increase, increasing its congestion window by c*R^2 packets
    per round-trip time, for per-flow round-trip time R and some global
    constant c, instead of TCP's Increase-by-One method of increasing
    the congestion window by one packet per round-trip time [F02]
    (Section 3.3), [F91].  In this document we argue that the benefits
    of RTT fairness would be outweighted by the inherent costs, for
    congestion control mechanisms that don't involve explicit feedback
    from routers.  We also give some suggestions of scenarios that could
    be used to explore some of the costs of congestion control
    algorithms designed to achieve RTT fairness.


1.  Introduction

    Discuss the various ways that congestion control mechanisms can be
    designed to give RTT fairness.

2.  Potential Benefits of RTT Fairness

    * Flows with longer RTTs receive more bandwidth that they would have
    otherwise.







Floyd                                               Section 2.  [Page 1]

INTERNET-DRAFT              Expires: May 2007              November 2006


3.  Potential Costs of RTT Fairness

    * Either the flows with RTTs of 10 ms and less are not aggressive
    enough to give satisfactory performance, or the flows with RTTS of
    500 ms and more are too aggressive to give satisfactory performance.

4.  Suggested Scenarios

    * Explore performance for a congested link where most flows have a
    round-trip time of 200 ms or more.

    * Then explore performance for a congested link where most flows
    have a round-trip time of 20 ms or less.

Informative References

    [F91] Floyd, S., Connections with Multiple Congested Gateways in
        Packet-Switched Networks Part 1: One-way Traffic, Computer
        Communication Review, Vol.21, No.5, October 1991, p. 30-47.

    [F92] Floyd, S., and Jacobson, V., On Traffic Phase Effects in
        Packet-Switched Gateways, Internetworking: Research and
        Experience, V.3 N.3, September 1992.

    [HK98] Henderson, T.R., E. Sahouria, S. McCanne, and R.H. Katz, "On
        Improving the Fairness of TCP Congestion Avoidance," IEEE
        Globecom `98, Nov. 1998.

Authors' Addresses

    Sally Floyd <floyd@icir.org>
    ICSI Center for Internet Research
    1947 Center Street, Suite 600
    Berkeley, CA 94704
    USA
















Floyd                                                           [Page 2]