*BSD News Article 34234


Return to BSD News archive

Xref: sserve comp.os.386bsd.questions:12305 comp.os.386bsd.misc:3171
Newsgroups: comp.os.386bsd.questions,comp.os.386bsd.misc
Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!bunyip.cc.uq.oz.au!munnari.oz.au!constellation!mimbres.cs.unm.edu!tesuque.cs.sandia.gov!lynx.unm.edu!nntp.sunbelt.net!udel!gatech!howland.reston.ans.net!agate!library.ucla.edu!csulb.edu!csus.edu!netcom.com!netcomsv!calcite!vjs
From: vjs@calcite.rhyolite.com (Vernon Schryver)
Subject: Re: Whats wrong with Linux networking ???
Message-ID: <CuBo2F.A5D@calcite.rhyolite.com>
Organization: Rhyolite Software
Date: Wed, 10 Aug 1994 13:56:39 GMT
References: <3256t1$rbn@ra.nrl.navy.mil> <327nj0$sfq@sundog.tiac.net> <328fn2$i9p@news.panix.com>
Lines: 33

In article <328fn2$i9p@news.panix.com> berke@panix.com (Wayne Berke) writes:
>bill@bhhome.ci.net (Bill Heiser) writes:
>
>>cmetz@sundance.itd.nrl.navy.mil (Craig Metz) writes:
>
>>>>- NFS was *slooow*
>>>	No arguing this one. Linux's NFS is still in need of serious work.
>
>>The *speed* of LINUX NFS isn't the real problem.  The reason it's so slow
>>is that by default it uses a 1K blocksize.  You can increase the rsize and
>>wsize to 8K, like Sun, and the performance improves dramatically.
>
>Could you explain why this should be the case?  Since 8K blocks will typically
>(eg. on an Ethernet) be fragmented by IP down to ~1K packets, why should these
>bigger blocks be an advantage.  If anything I would suspect that
>reassembly and retransmission costs would make the <MTU packets better.

If nothing else is going on, and if you have page-flipping and hardware
checksumming, the per packet TCP/IP or UDP/IP costs dominate.  IP
fragmenting and fragment reassembly is not hard, but it can matter a
lot, provided nothing else is going on.

NFS is often not network bound, even today sometimes not even on Ethernets.
NFS often runs 0.5 to 0.1 times as fast as `ttcp`.  Marshalling and
unmarshalling RPC-XDR args is far more than IP fragmenting or reassembly.
Then there is the real work of NFS, searching buffer and attribute
caches, figuring out block sizes, and on and on and on.  In other words,
the per-NFS packet processing is a lot of work.  It very much faster to
assemble a couple of IP fragments into one NFS packet than to have deal
with two NFS packets.


Vernon Schryver    vjs@rhyolite.com