[Bug 173541] load average 0.60 at 100% idle

bugzilla-noreply at freebsd.org bugzilla-noreply at freebsd.org
Thu Mar 3 14:51:46 UTC 2016


--- Comment #14 from smithi at nimnet.asn.au ---
The first 40% of the attached is a study of the "0.6" erroneous loadavgs
experienced also by others, shown in some detail over long periods at
virtually complete idle, plus a number of 100% CPU-bound tasks running.

This shows that the oft-mentioned "0.6" figure (seeming randomly variable
over a range of ~0.45 to ~0.75, with some outliers even beyond that) is an
ADDITIVE increment to real loadavg, for real 100% CPU loads up to here 11.

At 40% through this report are applied a pair of cpuset(1) commands, which
on this machine completely fix broken loadavgs, permanently until reboot.

I'll refrain from further speculation on what might cause this to happen.

The rest of the report illustrates the well-behaved nature of HPET-derived
loadavgs (both one-shot and periodic) at idle and at real loads of 1, 2, 4
and 11 CPU-bound tasks, after this particulat magic spell has been cast ..

Verbose dmesg is also attached.

You are receiving this mail because:
You are the assignee for the bug.

More information about the freebsd-bugs mailing list