[Tmrg] sending rate of a TCP flow

[Kostas.Pentikousis at vtt.fi (Pentikousis Kostas)]

Aydin:

| > hypothesis: When N long-lived TCP flows (with the same RTT and MSS
| > values) share a bottleneck link in a dumbbell topology (all the
| > edge links have the same delay and bandwidth), I expect each TCP
| > flow to have the same sending rate -- Assume no other cross
| > traffic in the network and drop-tail queues.
| >
| > Do you know any references (simulation-, emulation-, or real
| > experiment-based studies) that prove this hypothesis or the
| > opposite?
| 
| With good modern TCP this hypothesis has always held OK for me.

I second that, and I think it would be more interesting to see how SCTP
and DCCP perform.

| E.g., see figure 6 in [1].  In older TCPs without SACK we often did 
| find that there was in fact quite a disparity between similar 
| connections

[...]
| but in offers some evidence.

Some further evidence can be found in

K. Pentikousis and H. Badr, "An evaluation of TCP with Explicit
Congestion Notification", Annals of Telecommunications, vol. 59, no.
1-2, January-February 2004, pp. 170-198.  Available at
http://ipv6.willab.fi/kostas/?e2e&paper=ecn_eval

In particular the section titled "The case of global synchronization"
(pp. 15-16) provides some of the background related to your question.
You may also want to go through the ns-2 mailing list archives in the
early 2000s.

Interestingly, Wikipedia has an article on the topic:
http://en.wikipedia.org/wiki/TCP_global_synchronization

Best regards,

Kostas
http://ipv6.willab.fi/kostas