From owner-freebsd-hackers Wed Jan 31 01:54:03 1996 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.7.3/8.7.3) id BAA06297 for hackers-outgoing; Wed, 31 Jan 1996 01:54:03 -0800 (PST) Received: from haywire.DIALix.COM (root@haywire.DIALix.COM [192.203.228.65]) by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id BAA06246 for ; Wed, 31 Jan 1996 01:53:37 -0800 (PST) Received: (from news@localhost) by haywire.DIALix.COM (8.6.12/8.6.12) id RAA14754 for freebsd-hackers@freebsd.org; Wed, 31 Jan 1996 17:54:19 +0800 Received: from GATEWAY by haywire.DIALix.COM with netnews for freebsd-hackers@freebsd.org (problems to: usenet@haywire.dialix.com) To: freebsd-hackers@freebsd.org Date: 31 Jan 96 09:47:17 GMT From: peter@jhome.DIALix.COM (Peter Wemm) Message-ID: Organization: DIALix Services, Perth, Australia. References: <199601292254.AAA19955@plentium.clinet.fi>, <199601301746.LAA05539@brasil.moneng.mei.com> Subject: Re: Good news -- pipe stuff Sender: owner-hackers@freebsd.org Precedence: bulk jgreco@brasil.moneng.mei.com (Joe Greco) writes: >> One interesting note, I benchmark FreeBSD vs. Linux (and used to SVR4) >> regularly in order to evaluate places where performance might/should >> be improved. I am *brutal* to FreeBSD, but it is getting difficult to >> find places where it can easily be improved (some of the performance >> "nits" are due to differences in philosophy and not actual performance >> problems.) With the latest pipe improvements, I am running out of steam. >> Indeed, my goal is to "find" performance problems. If anyone has a "cache" >> of programs to show performance bottlenecks, please email them to me. >> They will be used to improve FreeBSD's performance, and if I don't >> do it, DG, BDE or someone else will work on the code. >> >> There is another layer of improvements that I have been thinking about, but >> those require more involved work, and I want to work on easier stuff right >> now :-). Brain vacation time :-). >> >> I would suggest taking a look into uptime benchmark. Make FreeBSD with >> 50-100 simultaneous users, WWW server, news server, ftp server and lots of >> nfs in the same machine to stay up for at least weeks in row, instead of >> days. Popularity of the results is guaranteed, and that is the benchmark >> professionals value the most. >My Web server was up 121 days before a power outage knocked it down, that is >pretty good given that the site record is (I believe) 135 days for a SunOS >4.1 box. >In my opinion my news server isn't as stable. I see about 1-2 week uptimes. >Since I've installed ccd and increased my alt.binaries partition size, I am >no longer running out of disk space on a regular basis and I haven't seen >that pesky "panic: free vnode isn't" message lately. Again, that is not a >guarantee that there's a relationship, but there is an apparent >correlation. Well, our news server (running an older -current kernel) used to fall over on an average of about 3 days. I took a chance and updated it (and several others) to -current as of about 11 Jan. It's not crashed since!! (uptime is about 18 days, a definate record for this machine!) I also turned MMAP support back on in innd, which previously used to make the crash rate worse, and now there's no noticable difference except innd runs faster. It's caught up on it's nntp backlog and is now has the second longest uptime of the site cluster (we had power problems recently too). This kernel was built a few days before John's last round of speedups. Cheers! -Peter >... Joe >------------------------------------------------------------------------------- >Joe Greco - Systems Administrator jgreco@ns.sol.net >Solaria Public Access UNIX - Milwaukee, WI 414/342-4847