*BSD News Article 70591


Return to BSD News archive

#! rnews 2805 bsd
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!news.rmit.EDU.AU!news.unimelb.EDU.AU!munnari.OZ.AU!news.hawaii.edu!news.uoregon.edu!vixen.cso.uiuc.edu!sdd.hp.com!swrinde!howland.reston.ans.net!math.ohio-state.edu!news.cis.ohio-state.edu!rutgers!utcsri!cs.toronto.edu!schenk
Newsgroups: gnu.misc.discuss,comp.os.linux.networking,comp.unix.bsd.netbsd.misc
From: schenk@cs.toronto.edu (Eric Schenk)
Subject: Re: TCP latency (Re: GNUseless Debates (Was Re: GNU software on a Linux system))
X-Nntp-Posting-Host: qew.cs.toronto.edu
Message-ID: <1996Jun11.004639.8574@jarvis.cs.toronto.edu>
X-Newsreader: NN version 6.5.0 #8 (NOV)
References: <1JWBCFZN@wuschel.ibb.schwaben.com> <4p7cne$2ri@linux.cs.Helsinki.FI> <4pa83i$2km@solutions.solon.com> <4pcsmv$l9p@master.di.fc.ul.pt> <4pd10g$dmq@solutions.solon.com> <4pfgbi$92l@master.di.fc.ul.pt>
Date: 11 Jun 96 04:46:39 GMT
Lines: 36
Xref: euryale.cc.adfa.oz.au gnu.misc.discuss:26261 comp.os.linux.networking:41271 comp.unix.bsd.netbsd.misc:3751

roque@master.di.fc.ul.pt (Pedro Roque Marques) writes:

>Peter Seebach (seebs@solutions.solon.com) wrote:
>: Pedro Roque Marques <roque@master.di.fc.ul.pt> wrote:
>: >RFC 1323 defines TCP options. If the linux box is acting as a router there
>: >is absolutly no chance it would change this fields unless you are using
>: >masquerading/transparent proxying. So the router being Linux or not has
>: >nothing to do with it.

>: But it does, by observation; if the BSD boxes are connected directly,
>: they get good speeds, if they are connected through a Linux box, using
>: the same pair of modems, and an otherwise-workable ethernet net between
>: one of them and the Linux box, we suddenly get *very* bad performance;
>: maybe 10 bytes/second.

>There is one issue that could get into play here: it's likely that the
>fact that options are present in packets makes header compression fail
>for every packet while still taking the performance cost of maintaining
>the connection states. Try switching ppp header compression off.

Ah. This tweaks a memory. I seem to recall that a bug in the VJ compression
code got fixed sometime in the 1.3.X development cycle. Also, the
sort of speed Peter is report is consistent with reports of problems
with VJ compression incompatabilities. [Mere failure to be able to do
header compression should not cause the performance hit that Peter
is seeing.]

In any case, I'd be interested to hear if either turning off VJ compression
or upgrading the Linux box to the 2.0 kernel fixes the problem.

-- eric

---------------------------------------------------------------------------
Eric Schenk                          www: http://www.cs.toronto.edu/~schenk
Department of Computer Science	               email: schenk@cs.toronto.edu
University of Toronto