Date: Thu, 18 Jan 1996 13:35:06 -0600 (CST) From: mailing list account <lists@argus.flash.net> To: rcarter@geli.com (Russell L. Carter) Cc: freebsd-hackers@freebsd.org Subject: Re: BSDvs Lxxxxx Flame.. Message-ID: <199601181935.NAA02998@argus.flash.net> In-Reply-To: <199601181259.EAA07365@geli.clusternet> from "Russell L. Carter" at Jan 18, 96 04:59:43 am
next in thread | previous in thread | raw e-mail | index | archive | help
In reply: > To: joerg_wunsch@uriah.heep.sax.de (Joerg Wunsch) > Subject: Re: BSDvs Lxxxxx Flame.. > Date: Thu, 18 Jan 1996 04:59:43 -0800 > From: "Russell L. Carter" <rcarter@geli.com> > > } As Marty Leisner wrote: > } > > } > > } > Can someone provide hard information about nfs serving capability > } > on identical hardware between linux and freebsd? > } > > } > I'm very disappointed at the nfs performance on linux...I've used > } > } No ``hard information'', but most linux guys admit that their NFS > } server is about the worst piece of the system. > } > } I've seen 800 KB/s being piped out of a FreeBSD 1.1.5.1 machine, going > } to a single SGI Indy. > } > > A good example how this sometimes works is disk performance. > About 9 months ago I started pointing out on the linux lists that > bonnie with a file size quite a bit larger than main memory size is > a good indicator of actual disk performance, and slowly that > discussion has gotten more realistic. (fewer "Zowwy! I get 15 MB/s out > of my EIDE drive on a 486!) > > The same needs to be done for the networking side. I'm quite partial > to netperf, myself. as for vm/drive performance, the best test i've come across is pitest from nasa/ames. i uploaded it to /pub/FreeBSD/incoming a long time ago, it's still there. pitest will do pi to an arbitrary number of digits, but requires the WHOLE number to be in core. raw cpu can be determined by using /usr/bin/time -l on it and setting the parameter MX to just barely fill all available core without going into swap, vm performance can be determined by adding one to that MX and thereby doubling the size of the pi [~50% of the pi into swap]. all of the documented results were done with MX=14 [benchmark number]. i've never tested linux with it, but freebsd is impressive since the vm merge was debugged in 2.0.5-R and fractionally better with 2.1.0-R, another 5% or so of speed improvement can be achieved by recompiling /usr/lib/*.a with -O2 -m486 -fomit-frame-pointer -fno-strength-reduce [disclaimer: make copies of all .o files in /usr/lib before re-installing /usr/lib after compile and then copy them back after installing, i don't want to hear "i can't get anything to compile and run anymore!"]. i expect linux's vm results would be as bad as it's nfs... btw: 15mb/s, must not have gotten out of the cache in the benchmark... Jim -- All opinions expressed are mine, if you | "I will not be pushed, stamped, think otherwise, then go jump into turbid | briefed, debriefed, indexed, or radioactive waters and yell WAHOO !!! | numbered!" - #1, "The Prisoner" jbryant@argus.flash.net - FlashNet Communications - Ft. Worth, Texas
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199601181935.NAA02998>