[e2e] A Framework for Window-based Congestion Control

[A.L.Narasimha Reddy]

We have taken a slightly different approach to this problem.
Our results are available on the web at

http://dropzone.tamu.edu/techpubs/2001/sumitha.pdf  -- a longer/thesis version

and at http://ee.tamu.edu/~reddy/sumitha.pdf -- shorter paper version.

The abstract of the work:

In this paper, we propose a new class of protocols called the ``Delayed 
Congestion Response Protocols'' where response to congestion in the network 
is deliberately delayed for some time interval, allowing applications to 
make necessary adjustments for impending loss in bandwidth. 
We provide the general framework for this class of protocols and examine 
three cases in particular. For these three cases we develop the analytical 
models and derive the conditions under which they can be
fair to TCP. We confirm the analysis through ns-2 based simulations. By 
showing that delayed congestion response is possible, we lay the ground work 
for the development of a new class of transport protocols which are capable 
of providing early warning to the application regarding an impending 
reduction in the sending rate. In addition, these protocols can be designed 
to have a smooth congestion response for multimedia applications. 

------------------included message-------------------------------


> From nishanth at cs.utexas.edu Tue Apr 23 10:27:25 2002
> From: "Nishanth R. Sastry" <nishanth at cs.utexas.edu>
> To: <end2end-interest at postel.org>
> cc: "Simon S. Lam" <lam at cs.utexas.edu>
> MIME-Version: 1.0
> X-AntiVirus: scanned by AMaViS 0.2.1
> Subject: [e2e] A Framework for Window-based Congestion Control
> X-BeenThere: end2end-interest at postel.org
> X-Mailman-Version: 2.0.1
> List-Help: <mailto:end2end-interest-request at postel.org?subject=help>
> List-Archive: <http://www.postel.org/pipermail/end2end-interest/>
> Date: Tue, 23 Apr 2002 10:24:07 -0500 (CDT)
> 
> 
> We would like to report a new mathematical framework called CYRF for
> designing congestion control protocols that converge to fairness and
> efficiency in the same sense as TCP (and under the same network condition
> assumptions as the Chiu-Jain proof for the convergence of TCP). We also
> show how to make these flows TCP-friendly.
> 
> As an experiment in using this framework, we have developed two protocols
> for streaming-media like applications:  LOG, where we balance the need for
> smoothness with the need for a fast response to network congestion
> indications; and SIGMOID, which is fair to TCP even in the presence of
> drop-tail queues (Smooth non-linear window based algorithms such as LOG
> and the binomial algorithms are TCP-friendly only with RED queues, which
> is a major problem for deployment in the current Internet) by giving up
> smoothness in return for a guaranteed minimum throughput.
> 
> This framework includes several commonly known protocols such as TCP,
> GAIMD and binomial congestion controls as special cases, and thus provides
> a common theoretical foundation for window based congestion control. A
> second possible application of this framework would use it to derive
> results common to all these protocols.
> 
> 
> A technical report with details is available for download from
> http://www.cs.utexas.edu/ftp/pub/techreports/tr02-09.ps.Z
> 
> Uncompressed pdf and ps versions of this and other recent technical
> reports on congestion control from our group are available from:
> http://www.cs.utexas.edu/users/lam/NRL/TechReports/
> 
> We'd greatly appreciate any comments or suggestions.
> 
> -------
> Sastry, Nishanth R. and Simon S. Lam. "CYRF: A Framework for Window-based
> Unicast Congestion Control." The University of Texas at Austin, Department
> of Computer Sciences. Technical Report TR-02-09. January 2002. 30 pages.
> 
> Abstract:
> This work presents a comprehensive theoretical framework of window based
> congestion control protocols called CYRF (for Choose Your Response
> Function) that are designed to converge to fairness and efficiency. We
> first derive a sufficient condition for convergence to fairness. Using
> this, we show how fair window increase/decrease policies can be
> constructed from suitable pairs of monotonically non-decreasing functions.
> We also give a new characterization of TCP-friendliness and simple rules
> for smooth CYRF flows to be TCP-friendly. Specific protocols that meet the
> needs of a given situation can be designed within this framework. We
> experimentally investigate two CYRF protocols for streaming media-like
> applications: LOG, a protocol suitable for applications that need to
> trade-off between smoothness and aggressiveness, and SIGMOID, which is
> designed to be "TCP-friendly" even in a network with drop-tail queues
> (unlike other non-linear window based protocols).
> 
> - Nishanth
> 
>