Date: Sun, 28 Oct 2007 15:29:02 +0200 From: Gunther Mayer <gunther.mayer@googlemail.com> To: Kris Kennaway <kris@FreeBSD.org> Cc: freebsd-questions@freebsd.org Subject: Re: CPU usage 100% but no process hogging CPU Message-ID: <47248E9E.8030203@googlemail.com> In-Reply-To: <47237628.9050302@FreeBSD.org> References: <47232945.10506@gmail.com> <47233334.8040005@FreeBSD.org> <47236DB8.4040403@googlemail.com> <47237628.9050302@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Kris Kennaway wrote: > Gunther Mayer wrote: > >> I don't see why my javavm, apache, postgres and/or radiusd would >> spawn such short lived processes. Come to think of it, I know radius >> might be doing just that, but how the heck would I go about finding >> out? top -H brings me no closer... > > Either increase the sample resolution or rule out other programs as > the cause. Increasing sample resolution just increases the CPU usage of top itself but gives no more information about the real culprits :-( But I managed to solve my problem by reverting to and upgrading my SMP kernel. The system's back now with 15 minute load averages around 0.05 which is where it should be. What really remains a puzzle is what caused the bottleneck on my system running the GENERIC single CPU system. I actually went back to my daily cron emails in which it states that load average at 5am increased from around 0 to 1.0 somewhere between day 3 and day 4 after rebooting into freebsd-update's erroneously downloaded GENERIC kernel. It stayed around 1.0 until this morning when I reverted and rebooted. On even closer inspection, before doing freebsd-update I had an uptime of 202 days on that box. Even then there are some days when load average was 1.0 at 5am, I think the real culprit is my radiusd which probably sometimes spawns threads that chow 100% CPU until killed somehow. Having only one CPU available would of course exacerbate such a problem. Seeing that performance is back 100% I think I will wait for 7.0 to supercharge my system's threads. Anyways, thanks for the help Kris. Gunther > >> I see that my java is using no less than 26 threads, thread usage not >> showing up might well be the problem. >> >> Interesting point you make about libthr. I had no idea about the >> different threading options available on 6.x and did some reading up >> on it but there's very little official documentation or >> recommendations about it. Upon investigation it turns out that all my >> core daemons (httpd, postgres, radiusd and java) are linked against >> /lib/libpthread.so.2 which afaict after limited reading is what you >> refer to as libkse. Is that correct? > > Yes. > >> So I'm tempted to use libmap.conf to switch to libthr for all >> instances of libpthread, though I'm put off by some very recent >> reports (http://roy.marples.name/node/332) that this can cause some >> nasty problems. Do you think that's cause for concern? > > That bug report makes no sense to me. If they are using libmap > correctly then *all* libpthread references are remapped and if there > are missing symbols they will cause failure from the dynamic linker > when the process is first executed, not random crashes during operation. > >> Would you mind sharing your libmap.conf and/or symlink setup with the >> list as well as your experiences with libthr? > > See libmap.conf(5). > >> Then again, as a first step I should really get my SMP kernel back as >> a first step... > > That will surely help. > > Kris >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?47248E9E.8030203>