lkml.org 
[lkml]   [1997]   [May]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: TCP/IP works better in 2.1.37 than 2.0.30
Date

Jari Soderholm <kernel@atmweb.ntc.nokia.com> writes:
>>I just switched from 2.0.30 to 2.1.31 and there seems to
>>be less magical waits in the middle of connetions which
>>have have been happening as long as i rememember.
>
>Ouch !
>
>I was trying to say that
>
>TCP/IP works better in 2.1.37 that 2.0.30

This is a bit odd, the two TCP implementations should have pretty much
identical behavior, at least until you turn on some of the new optimization
code or the RFC1323 stuff.

If you are just running 2.1.37 without mucking with the settings in
/proc/sys/net/ipv4, then I can only think of two differences.
First, in 2.0.30 there are a few bugs that have since been fixed in
pre-2.0.31 plus a few more patches from me. In particular there could
be slightly more stalling because of the way that 2.0.30 picks TCP window
sizes, but I would be a bit surprised by that.
The only other thing I can think of is actually a bug in 2.1.37.
Currently it doesn't do delayed ACKs properly, which results in twice
as many ACKs being sent as would otherwise be the case. Now, when
packets are getting lost occasionally, this bug actually improves
performance a bit... on the other hand it causes other problems, so
we'll be fixing it anyway. I'll be interested to hear how 2.0.31 and
say 2.1.39 (or whenever I get around to fixing that delayed ACK bug)
compare.

--
Eric Schenk www: http://www.dna.lth.se/~erics
Dept. of Comp. Sci., Lund University email: Eric.Schenk@dna.lth.se
Box 118, S-221 00 LUND, Sweden fax: +46-46 13 10 21 ph: +46-46 222 96 38



\
 
 \ /
  Last update: 2005-03-22 13:39    [from the cache]
©2003-2014 Jasper Spaans. Advertise on this site