[e2e] TCP un-friendly congestion control

[Craig Partridge]

In message <200306062025.h56KPvdV015271 at ginger.lcs.mit.edu>, "J. Noel Chiappa" 
writes:

>I would suspect that in practise, the first aspect I mentioned (two different
>kinds of loss signal getting mixed up) is probably an even bigger problem
>than the other one (the difficulty of using such a small signal for
>feedback).

Hi Noel:

I don't think there's a lot of data on this point.

The only work I know of immediately is some work done at BBN on explicit error
notification.

One of the experiments done was to assume that you had perfect knowledge
of what loss was caused by error and what was loss was caused by congestion
(so error-loss didn't cause a congestion response).  This was simulation of
course -- and so it was possible to do a test where we went even farther
and notified the source the instant of the error-related loss (so we
cut out the control loop effect) so that it could retransmit immediately.

The results were somewhat deflating.  You could see an improvement in
throughput, but it wasn't very impressive.  At some point we need to write
up the results in a paper but there is a tech report on-line at
http://www.ir.bbn.com/documents/techmemos/index.html (look for BBN-TR-8333).

Thanks!

Craig