From owner-freebsd-hackers Mon Mar 15 9: 4: 7 1999 Delivered-To: freebsd-hackers@freebsd.org Received: from cygnus.rush.net (cygnus.rush.net [209.45.245.133]) by hub.freebsd.org (Postfix) with ESMTP id DD45C14BD6; Mon, 15 Mar 1999 09:03:59 -0800 (PST) (envelope-from bright@rush.net) Received: from localhost (bright@localhost) by cygnus.rush.net (8.9.3/8.9.3) with SMTP id MAA29998; Mon, 15 Mar 1999 12:08:14 -0500 (EST) Date: Mon, 15 Mar 1999 12:08:12 -0500 (EST) From: Alfred Perlstein To: Eivind Eklund Cc: freebsd-hackers@FreeBSD.ORG Subject: tuneing, was: Re: Has anybody used Postmark for file system benchmark In-Reply-To: <19990315153735.B98270@bitbox.follo.net> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG On Mon, 15 Mar 1999, Eivind Eklund wrote: > On Mon, Mar 15, 1999 at 08:27:04AM -0000, Yusuf Goolamabbas wrote: > > Postmark is available at > > > > http://www.netapp.com/technology/level3/3022.html > > > > I am currently trying it out on various FreeBSD 3.1-STABLE boxes (It > > compiled cleanly). Thought I would check with others if they had any > > experience with this. According to the paper , UFS on Solaris 2.5 > > sucks so I am seeing what FreeBSD (and possibly softupdates can do) > > > > Hope to provide some results soon. I shall be also trying on Linux > > boxes (On Linux, I am getting the following error) > > Remember that the default tuning on the FreeBSD boxes and the Linux > boxes are different; this has to be reflected if you're going to > attempt to do a fair benchmark. The different modes are > > Linux FreeBSD > Fully synchronous sync sync > "Synchronous" metadata, > async data N/A (default mode) > Fully async (unordered > metadata writes)[1] default async > Tracking dependencies N/A "soft updates" > > [1] This mode is highly disfunctional if your data is critical. The > normal file system invariants (e.g, that data you write end up in a > file you own) are not honoured across unscheduled reboots. It is very > useful for news spools etc, of course. > > Comparing the Linux default to the FreeBSD default is pretty > uninteresting; the tradeoffs are totally different. IMO, the most > interesting compares are Linux sync against FreeBSD > anything-but-async, and Linux async vs FreeBSD async. > > When you're giving out results, please also include information about > the exact hardware platform - a hardware platform that caches will > tend to skew the results. I think that it's very important that a file, perhaps BENCH.TXT or equiv explain these issues. A lot of performance options are turned off to help the install actually work, just because it installs this way doesn't mean it's tuned. Several of my friends use FreeBSD but aren't on these lists and don't have the time/knowledge to look into /usr/src/sys to tune things. Things to consider: pointers to: softupdates, turning on DMA/multiblock for IDE, removing the 386/486 options in the kernel increasing MAXUSERS/NMBCLUSTERS for heavily loaded machines I started to put togther a "tuning freebsd" page a year ago, it sucks, it's out of date, but you can check it out here: http://www.genx.net/~bright/unix/optimizing.html don't laugh, i haven't touched the pages in over a year... :) I think it's so important that a link off the main freebsd page would be a good idea (not to my page) but something done and kept updated like src/UPDATING and perhaps a dialog at the end of sysinstall: "Your install is complete, please visit http://www.freebsd.org/tuning for information on improving your machine's performance" I know people using freebsd longer than I have who didn't know about IDE tuning until I told them about it. Doesn't it suck when you see a FreeBSD benchmark and you see the quote "we used the default settings"? -Alfred > > Eivind. > To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message