From owner-freebsd-stable Thu Jun 11 14:45:53 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id OAA07568 for freebsd-stable-outgoing; Thu, 11 Jun 1998 14:45:53 -0700 (PDT) (envelope-from owner-freebsd-stable@FreeBSD.ORG) Received: from hobbes.saturn-tech.com (drussell@drussell.internode.net [198.161.228.154]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id OAA07498 for ; Thu, 11 Jun 1998 14:45:32 -0700 (PDT) (envelope-from drussell@saturn-tech.com) Received: from localhost (drussell@localhost) by hobbes.saturn-tech.com (8.8.4/8.8.2) with SMTP id PAA06334; Thu, 11 Jun 1998 15:44:52 -0600 (MDT) Date: Thu, 11 Jun 1998 15:44:52 -0600 (MDT) From: Doug Russell To: "Viren R. Shah" cc: stable@FreeBSD.ORG Subject: Re: NFS server performance (was: NFS performance benchmarks?) In-Reply-To: <199806111748.NAA08167@fault.rstcorp.com> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-stable@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG On Thu, 11 Jun 1998, Viren R. Shah wrote: > We tried iozone, and the results were bad (as shown below), which is > why I was trying to see if there were other benchmarks that validated > the iozone results, or not. > We are trying to compare the relative performances of a solaris NFS > server vs. a FreeBSD NFS server (running 2.2.6 late BETA): > > IOZONE: auto-test mode > > > FreeBSD Server Solaris Server > ====================== ================== > MB reclen bytes/sec bytes/sec bytes/sec bytes/sec > written read written read > 1 512 51737 7117484 486685 7139923 > 1 1024 48598 11162427 472185 12754985 > 1 2048 53363 16469678 465276 19026272 > 1 4096 57415 25860745 492552 26323628 > 1 8192 56089 30133306 490601 33321942 > 2 512 52701 6225125 478547 7336163 > 2 1024 53642 10811569 483381 11907853 > 2 2048 46550 16377591 461267 18763777 > 2 4096 55467 22791909 489909 25800971 > 2 8192 52679 30147920 474295 22596409 > 4 512 51664 5246158 484840 6519525 > > > Both the iozone tests were run from a SunOS NFS client. As you can see > the write performance is an order of magnitude worse. The local iozone > results for both servers were comparable (though the FreeBSD box had > slightly worse performance) There must be some time of flakiness occuring here. (Perhaps some strange incompatibility due to the SunOS client??) I have used iozone to test NFS performance before, but I did a quick test here to check that nothing has broken recently. With an AM486DX4/120 client running 2.2.6-STABLE from a couple weeks ago with an old NE2000 clone, mounting /usr from an AM5x86/150 running ancient 2.2-BETA with a 3c905 in 10 Mbps mode, I get: Writing the 20 Megabyte file, 'iozone.tmp'...46.320312 seconds Reading the file...25.390625 seconds IOZONE performance measurements: 452749 bytes/second for writing the file 825955 bytes/second for reading the file iozone auto results are similar, but when you get to the really small file sizes, various caches seem to speed things up abnormally, so this is probably a better result. Obviously my read speeds are limited due to the 10mbps link and NE2000, but this seems to be more the expected result. I may play with this a little later on some newer boxes and see what happens. I don't have any Sun boxes on my networks, and rather than drag one over from the other end of the building and reconfiguring it, perhaps someone else can shed some light on this? :) Later...... To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message