Date: Wed, 29 Sep 2010 16:27:29 -0700 (PDT) From: Don Lewis <truckman@FreeBSD.org> To: avg@icyb.net.ua Cc: stable@FreeBSD.org, sterling@camdensoftware.com, freebsd@jdc.parodius.com Subject: Re: CPU time accounting broken on 8-STABLE machine after a few hours of uptime Message-ID: <201009292327.o8TNRTPa077418@gw.catspoiler.org> In-Reply-To: <4CA3010D.9080909@icyb.net.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
On 29 Sep, Andriy Gapon wrote: > on 29/09/2010 11:56 Don Lewis said the following: >> I'm using the same kernel config as the one on a slower !SMP box which >> I'm trying to squeeze as much performance out of as possible. My kernel >> config file contains these statements: >> nooptions SMP >> nodevice apic >> >> Testing with an SMP kernel is on my TODO list. > > SMP or not, it's really weird to see apic disabled nowadays. I tried enabling apic and got worse results. I saw ping RTTs as high as 67 seconds. Here's the timer info with apic enabled: # sysctl kern.timecounter kern.timecounter.tick: 1 kern.timecounter.choice: TSC(800) ACPI-fast(1000) i8254(0) dummy(-1000000) kern.timecounter.hardware: ACPI-fast kern.timecounter.stepwarnings: 0 kern.timecounter.tc.i8254.mask: 65535 kern.timecounter.tc.i8254.counter: 53633 kern.timecounter.tc.i8254.frequency: 1193182 kern.timecounter.tc.i8254.quality: 0 kern.timecounter.tc.ACPI-fast.mask: 16777215 kern.timecounter.tc.ACPI-fast.counter: 7988816 kern.timecounter.tc.ACPI-fast.frequency: 3579545 kern.timecounter.tc.ACPI-fast.quality: 1000 kern.timecounter.tc.TSC.mask: 4294967295 kern.timecounter.tc.TSC.counter: 1341917999 kern.timecounter.tc.TSC.frequency: 2500014018 kern.timecounter.tc.TSC.quality: 800 kern.timecounter.invariant_tsc: 0 Here's the verbose boot info with apic: <http://people.freebsd.org/~truckman/AN-M2_HD-8.1-STABLE-apic-verbose.txt>; I've also experimented with SMP as well as SCHED_4BSD (all previous testing was with !SMP and SCHED_ULE). I still see occasional problems with SCHED_4BSD and !SMP, but so far I have not seen any problems with SCHED_ULE and SMP. I did manage to catch the problem with lock profiling enabled: <http://people.freebsd.org/~truckman/AN-M2_HD-8.1-STABLE_lock_profile_freeze.txt>; I'm currently testing SMP some more to verify if it really avoids this problem.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201009292327.o8TNRTPa077418>