Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 9 Nov 2001 14:47:11 -0700
From:      Nate Williams <nate@yogotech.com>
To:        stable@FreeBSD.org
Cc:        jayanth@FreeBSD.org
Subject:   TCP NewReno causing wild performance fluctuations
Message-ID:  <15340.20191.441648.662113@caddis.yogotech.com>

next in thread | raw e-mail | index | archive | help
We've been using FreeBSD boxes as our reference ftp servers at work, and
because of the recent security issues, I've went through and updated a
number of our public boxes from 3.x -> 4.4.

However, we're seeing significant variations in ftp throughput from
boxes that are directly connected on the lan segment.  If I disable the
New Reno code, things go back to normal (ie; no fluctuations).

I went through the logfiles, and it turns out there has been one fix to
the code that hasn't been merged into stable.

Rev1.139 in tcp_input.c has not (yet) been merged into -stable.

revision 1.139
date: 2001/08/29 23:54:13;  author: jayanth;  state: Exp;  lines: +3 -1
when newreno is turned on, if dupacks = 1 or dupacks = 2 and
new data is acknowledged, reset the dupacks to 0.
The problem was spotted when a connection had its send buffer full
because the congestion window was only 1 MSS and was not being incremented
because dupacks was not reset to 0.

Obtained from:          Yahoo!

I hand-merged this change back onto my box, installed a new kernel, and
now in my *very* minor testing things appear to be more normal.  My
question is should this fix be merged into stable, and would the lack of
this bugfix explain my performance results?

Unfortunately, I don't haven't gotten any tcpdump outputs to analyze
what was going on, but I didn't want this bugfix to slip through the
cracks.



Nate



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?15340.20191.441648.662113>