9.3 Process Averages
Ian Smith
smithi at nimnet.asn.au
Sun Oct 19 16:51:11 UTC 2014
On Sun, 19 Oct 2014 08:44:44 -0700, Michael Sierchio wrote:
> Is device polling enabled?
Nope.
> The load average is a global number of runnable-but-waiting processes, but
> you should divide by the number of cores to get something more intuitively
> meaningful
Yeah it's supposed to be. I just applied the fix referred to in:
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=173541
itself referred from this forum post:
https://forums.freebsd.org/threads/high-load-average-with-idle-state.38757/
Namely, setting '# sysctl kern.eventtimer.timer=LAPIC' (previously HPET)
on my (still) 9.3-PRERELEASE system, the load averages shown immediately
started to drop from the previously shown 0.6 or so, after a minute or
two to 0.10, 0.30, 0.45, and now a few minutes later to:
last pid: 97253; load averages: 0.14, 0.09, 0.12 up 104+00:06:03 03:31:07
87 processes: 3 running, 83 sleeping, 1 waiting
CPU: 0.4% user, 0.0% nice, 3.4% system, 0.0% interrupt, 96.1% idle
Mem: 468M Active, 787M Inact, 482M Wired, 9680K Cache, 202M Buf, 116M Free
Swap: 2048M Total, 153M Used, 1895M Free, 7% Inuse
ie, about what I'd expect from an essentially idle system.
Thanks to Alex, whose message just arrived here; I'd read it on the list
webpage before following the above.
(Alex, that PR is there if you remove the 'kern/' ie just as above ..
perhaps to do with the change to Bugzilla?)
Whether using LAPIC instead of HPET has any consequences here remains to
be seen.
cheers, Ian
More information about the freebsd-questions
mailing list