[Iccrg] Heresy recapped
dirceu_cavendish at yahoo.com
Thu Apr 17 18:44:03 BST 2008
I will not dwell on VJ "intention" regarding fairness issues. However, from a control theoretical perspecive, it is a fact that a ON-OFF type of rate regulation embedded in the AIMD feature of TCP CA causes bottleneck bandwidth resharing, converging* to similar transmission rates.
*I should be careful about the term convergence, as theoreticcally an ON-OFF controller never converges, but in steady state oscillates around a "fair" point. The oscillation band can be related with teh session rtt...
----- Original Message ----
From: "Eddy, Wesley M. (GRC-RCN0)[VZ]" <wesley.m.eddy at nasa.gov>
To: Michael Welzl <michael.welzl at uibk.ac.at>; iccrg <iccrg at cs.ucl.ac.uk>
Sent: Thursday, April 17, 2008 6:11:10 AM
Subject: RE: [Iccrg] Heresy recapped
>From: iccrg-bounces at cs.ucl.ac.uk
>[mailto:iccrg-bounces at cs.ucl.ac.uk] On Behalf Of Michael Welzl
>Sent: Thursday, April 17, 2008 1:38 AM
>>I will just point out that fairness is not a goal of VJ CC; avoiding
>>packet loss is.
>>The inputs to VJ CC are "some bytes were not lost" (an ACK) and "some
>>bytes were probably lost" (an RTO or 3-dupack). They are not "someone
>>is sending faster than me" or "I'm sending faster than someone else",
>>or any other metric or inference useful for acheiving any form of
>>fairness, because that is not its goal.
>Now that ain't true. The choice of AIMD in the Cong. Avoid & Control
>paper is based on the Chiu/Jain paper (this paper is cited in Van's),
>where AIMD was found to be the only control leading to efficiency
>*and fairness* among the four simple linear ones that they looked at.
The message at:
only briefly mentions fairness, and actually labels it as something
for the router to decide, not the end hosts.
"While algorithms at the transport endpoints can insure the network
capacity isn't exceeded, they cannot insure fair sharing of that
capacity. Only in gateways, at the convergence of flows, is there enough
information to control sharing and fair allocation."
Both references describe the selection of AIMD without reference to
fairness, but specifically for stability.
Iccrg mailing list
Iccrg at cs.ucl.ac.uk
Be a better friend, newshound, and
know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Iccrg