Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!ns.saard.net!hakea.adelaide.edu.au!news.cs.su.oz.au!metro!metro!munnari.OZ.AU!news.ecn.uoknor.edu!news.wildstar.net!news.ececs.uc.edu!cloudbreak.rs.itd.umich.edu!newsxfer3.itd.umich.edu!news.bbnplanet.com!su-news-hub1.bbnplanet.com!newsfeed.internetmci.com!demos!news.uni-stuttgart.de!news.urz.uni-heidelberg.de!rz.uni-karlsruhe.de!not-for-mail From: uk1o@rzstud2.rz.uni-karlsruhe.de (Felix Schroeter) Newsgroups: comp.os.linux.misc,comp.os.linux.networking,comp.os.linux.setup,comp.unix.bsd.misc,comp.os.ms-windows.nt.advocacy,comp.os.os2.advocacy Subject: Re: Linux vs whatever Date: 4 Feb 1997 18:33:50 +0100 Organization: University of Karlsruhe, Germany Lines: 40 Message-ID: <5d7rtu$ao9@rzstud2.rz.uni-karlsruhe.de> References: <32DFFEAB.7704@usa.net> <5coib1$jvv@cynic.portal.ca> <5cov5p$7hu@lynx.dac.neu.edu> <5cphaj$qvg@cynic.portal.ca> NNTP-Posting-Host: rzstud2.rz.uni-karlsruhe.de Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 8bit Xref: euryale.cc.adfa.oz.au comp.os.linux.misc:156200 comp.os.linux.networking:67291 comp.os.linux.setup:95528 comp.unix.bsd.misc:2247 comp.os.ms-windows.nt.advocacy:51912 comp.os.os2.advocacy:265726 Hello! In article <5cphaj$qvg@cynic.portal.ca>, Curt Sampson <cjs@cynic.portal.ca> wrote: >[... thread shifted from BSD vs Linux code quality to > BSD vs Linux networking performance ...] >My NFS evidence is available at ><http://www2.portal.ca/~cjs/computer/benchmark>. So far all of >the stats I've seen from any Linux user have been far below these >read benchmarks on any class of machines, and below the write >benchmarks on machines with similar CPU power. Hmmm. IMHO the Linux NFS drivers are broken, but that has not much to do with the Linux TCP/IP stack. A test between a Linux box (NFS client, 486/133, 64 MB RAM, AH2940U, Quantum Fireball) to my box (OpenBSD, NFS server, P200, 64MB RAM, AH2940UW, IBM DORS ... wide HD) showed the following result: 2 MB *write*, 60 seconds, both boxen nearly w/o load except the NFS transfer. I used tcpdump to find out, what happened: The Linux NFS clients used 1 KB (!) NFS write requests, the server executed the requests, including the fsync() as required by the NFS specs (which slowed the writes down), *then* sent the acknowledges. The 1KB write size was hardcoded in the Linux (2.0.something) NFS client implementation. But slow NFS performance is *no* evidence for slow TCP/IP performance. The Linux NFS implementation is much more broken than its TCP/IP implementation can ever be :-/ And on modern boxen, you can't see the difference of the two TCP/IP implementations any more, unless you are using something faster than 10 MBit Ethernet. Regards, Felix.