From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: <9ab217670704180959hf2e7760pc09e2a43013a724@mail.gmail.com> Date: Wed, 18 Apr 2007 12:59:13 -0400 From: "Devon H. O'Dell" To: "Fans of the OS Plan 9 from Bell Labs" <9fans@cse.psu.edu> Subject: Re: [9fans] syscall spikes In-Reply-To: <20070418165314.CC74B1E8C26@holo.morphisms.net> MIME-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Content-Disposition: inline References: <9ab217670704170715i1e62ed48k9817177886add100@mail.gmail.com> <20070418165314.CC74B1E8C26@holo.morphisms.net> Topicbox-Message-UUID: 4c1f7242-ead2-11e9-9d60-3106f5b1d025 2007/4/18, Russ Cox : > > I've been having syscall spikes. It's really weird. There will be > >>40,000 in 1 second, (according to stats(8)) and it appears at a > > fairly regular interval. I've tried running kprof, but I'm not sure > > that it's running a global profile. Namely, when I come out of it, I > > have tons of time spent in ``halt'', with very, very few (<100) > > syscalls. Any ideas what I can do to find the culprit? It's gapping > > sound playback, which is really annoying. > > You figured out that this is caused by two timesyncs. > > However, it might not be caused by the timesyncs actually > issuing that many system calls as much as fighting over > what time means. So probably you're not having 40,000 > system calls but in fact the clock is being slowed/stopped > in the kernel, which makes the devaudio sleep take too long. Why would stats(8) say that there's anywhere between 40000 and (the highest I've seen) 80000 syscalls in that second? Is its notion of the second screwed up that much due to the contention? > Russ --dho