On Sat, Sep 12, 2015 at 12:31:16AM -0500, Bruce Dubbs wrote: > Ken Moffat wrote: > >On Fri, Sep 11, 2015 at 05:59:20PM -0500, William Harrington wrote: > >>On Fri, 11 Sep 2015 21:08:16 +0100 > >>Ken Moffat <[email protected]> wrote: > >> > >>>FAIL: rt/tst-mqueue5 > >>> > >>>Comparing my i3 build (LFS as at 1st September) I only saw > >>>proteccted1{a,b} and getaddrinfo{4,5} - I suspect all of the others > >>>are because I'm using AMD (yeah, I'm paranoid :) The > >>>double-vlen{2,4} do not ring any bells, but I think I've seen > >>>-double and -idouble on my other AMD box. > >>> > >>>As with all tests - they probably mean nothing on their own ;) > >>> > >>>ĸen > >>Message queuing is tested and the output is at rt/tst-mqueue5.out > >> > >>The test is time sensitive and may fail while using the system or the > >>system is slow. > > >Thanks. In my case, the host system was running X with 3 or 4 terms > >(urxvt) open, and probably running xscreensaver during the tests. > >The box had probably throttled back to its lowest cpufreq. ISTR > >that my AMD boxes drop to lower frequencies than my i3, and probably > >do less per Hz. > > I can understand the screensaver using cycles, but if it's running the > tests, it's curious why the cpu would slow down. > > -- Bruce > I'm using the ondemand governor - when a core is under-loaded, it slows down. If one test is running, that *one* core might be flat out (3700 MHz on this machine - I'm also seeing 3000, 2400, 1700).
BUT: the kernel will distribute work among the CPUs - I believe that is intended to even out the core heating. So, when the activity moves to a different CPU, that CPU has to be speeded up if there is very little work. ĸen -- Il Porcupino Nil Sodomy Est! (if you will excuse my latatian) aka "The hedgehog song" -- http://lists.linuxfromscratch.org/listinfo/lfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
