[e2e] Performance evaluation of high speed TCPs

[Saverio Mascolo]

hi,

we have shown effect of reverse traffic on TCP Vegas, New Reno and Westwood+ 
in an ACM CCR paper (april 2004)
you can find here:
http://193.204.59.68/mascolo/tcp%20westwood/ccr_v31.pdf

historically we were "forced" to look at reverse traffic because congestion 
on the ack path (and ack compression)  was provoking bad behaviour of 
Westwood TCP, which is also shown in the paper above. the reason was 
bandwidth overestimate because we were computing bandwdith samples every ack 
and smaples were aliased (see 
http://193.204.59.68/mascolo/recent_papers/Grieco_Mascolo_3621_QoSIP03.pdf )

other result was that  Vegas TCP is heavily affected by reverse traffic 
because of queuing on reverse path. similar sensitivity to reverse traffic 
have been found with FAST TCP in a paper appearing at pfldnet06 and also in 
tests done by les cottrel.

best,
saverio

----- Original Message ----- 
From: "Injong Rhee" <rhee at eos.ncsu.edu>
To: <end2end-interest at postel.org>
Cc: "'Long Le'" <le at cs.unc.edu>; <rhee at ncsu.edu>; "'Yusung Kim'" 
<yskim at cosmos.kaist.ac.kr>; "'Lisong Xu'" <xu at cse.unl.edu>
Sent: Monday, January 16, 2006 11:12 PM
Subject: [e2e] Performance evaluation of high speed TCPs


> We have just finished our first series of performance evaluation of 6
> different high TCP variants that have been recently proposed. We have
> a report on the results below:
>
> http://www.csc.ncsu.edu/faculty/rhee/export/bitcp/asteppaper.htm
>
> This report has been long overdue and can be considered as our
> response to earlier test reports by other folks publicized in several
> mailing lists.  What is unique about our report is that we emphasize
> on the importance of background traffic in experimental evaluation of
> congestion control protocols. The results we got with background
> traffic are quite different from what are reported in one of the
> earlier reports. It is not unknown that background traffic makes
> difference in protocol performance, but it is unknown how they make
> difference and what properties may change.
>
> Please view the report and follow the embedded links in the report to
> get more information about experiment. The performance pictures in the
> report are a little blurred due to resizing. But if you follow the
> html links associated with each picture you can get a better picture
> of it and more information about it.
>
> While we're preparing this report, we found that there are many
> testing scenarios that we need to cover, yet it is difficult to come
> up with a meaning scenario that could change protocol behaviors in a
> meaningful way. So we would like to solicit your opinion on types of
> testing scenarios you want to see. Also if you have other protocols
> that you like to see it being tested in the same environment, please
> let us know as well.
>
> Thanks
> Injong
> ----
> Injong Rhee
> Associate Professor
> Department of Computer Science
> North Carolina State University
> http://www.csc.ncsu.edu/faculty/rhee
>
>