*BSD News Article 34113


Return to BSD News archive

Xref: sserve comp.os.386bsd.questions:12236 comp.os.386bsd.misc:3122
Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!bunyip.cc.uq.oz.au!munnari.oz.au!constellation!news.uoknor.edu!ns1.nodak.edu!netnews.nwnet.net!pnl-oracle!osi-east2.es.net!cronkite.nersc.gov!dancer.ca.sandia.gov!overload.lbl.gov!agate!howland.reston.ans.net!europa.eng.gtefsd.com!newsxfer.itd.umich.edu!gumby!andrews-cc!gillham
From: gillham@andrews.edu (Andrew Gillham)
Newsgroups: comp.os.386bsd.questions,comp.os.386bsd.misc
Subject: Re: Whats wrong with Linux networking ???
Date: 9 Aug 1994 23:03:39 GMT
Organization: Andrews University
Lines: 36
Message-ID: <32920b$m5n@orion.cc.andrews.edu>
References: <Cu107E.Mz3@curia.ucc.ie> <3256t1$rbn@ra.nrl.navy.mil> <327nj0$sfq@sundog.tiac.net> <328fn2$i9p@news.panix.com>
NNTP-Posting-Host: edmund.cs.andrews.edu

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.

I would say the difference is the 'layers' that are involved.
At what layer in the OS is the IP packet fragged versus at what layer
is a 1k NFS read taking place?  How many system calls or context
switches are involved in making a NFS write call? (both ends)
Seems to me that if the upper layer asks to write 8k, jumps through
the system call hoops to get to the IP transmit layer, it only does that
once for 8k, instead of around 5 times.  Then the IP layer fragments it
and sends it.  Someone else would have to comment on whether it can send
the fragmented (to ~1500 bytes?) without context switching or making
many system calls.  Also I would imagine there is connection info, etc 
that each nfs write packet would duplicate.

-Andrew
-- 
#!/bin/sh - ==============================================
echo "Andrew Gillham                 gillham@andrews.edu"
echo "Winix Hacker"
#=========================================================