Consistently "high" CPU load on 10.0-STABLE

[Jeremy Chadwick]

On Sun, Jul 20, 2014 at 04:36:10PM -0700, hiren panchasara wrote:
> On Sat, Jul 19, 2014 at 11:24 PM, Jeremy Chadwick <jdc at koitsu.org> wrote:
> > (Please keep me CC'd as I'm not subscribed to freebsd-stable@)
> [skip]
> >
> > I do have DTrace enabled/built on this box but I have absolutely no clue
> > how to go about profiling things.  For example maybe output of this sort
> > would be helpful (but I've no idea how to get it):
> >
> > http://lists.freebsd.org/pipermail/freebsd-stable/2014-July/079276.html
> 
> You can probably use hotkernel or something similar?
> http://www.freebsd.org/doc/handbook/dtrace-using.html

Yeah, I've given that a try.  I think all that script does, in layman's
terms, is count the number of occurrences of a kernel function/symbol
being called.  If there was a single kernel function being called
repeatedly in excess then it would show up there, but if it's more like
a mutex being held for a long time I don't think hotkernel would show
this.  Meaning: it doesn't actually indicate "how much" time something
took, just the number of times something was run.  I think this was why
Adrian was advocating KTR.

procsystime might be more useful if I had reason to think it was a
userland process, but the jury's still out on that one (I haven't seen
anything userland show up in top that would indicate CPU usage, which is
why I'm more suspicious of the kernel).

-- 
| Jeremy Chadwick                                   jdc at koitsu.org |
| UNIX Systems Administrator                http://jdc.koitsu.org/ |
| Making life hard for others since 1977.             PGP 4BD6C0CB |