If I recall correctly, we'll end up using vmem_xalloc/vmem_xfree for
allocations larger than the arena quantum cache, or for allocations
via memalign (and presumably posix_memalign). I suspect this means
that most of your allocations are taking the slower uncached path.
That said, even if the above is correct (and I would check it), it
seems not particularly related to your actual problem, since that
would burn _user_ time, and you're chasing down what you feel is an
excess of system time.
Presumably there's a 'hotkernel.d', or a simple oneline:
profile-997hz / execname == "magick" / { @[stack()] = count(); }
To give you an idea where you're spending your system time. You also
appear to have slept 2 seconds in the example you gave, so you may
want to look into whether that was just being preempted, or I/O or ...
Also, an example not so wallclockily short and subject to outside
influence (other activity on the system? zfs txg syncing?) could be
helpful. I'm always nervous of (what amount to) short-running
benchmarks.
-- Rich
-------------------------------------------
illumos-discuss
Archives: https://www.listbox.com/member/archive/182180/=now
RSS Feed: https://www.listbox.com/member/archive/rss/182180/21175430-2e6923be
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21175430&id_secret=21175430-6a77cda4
Powered by Listbox: http://www.listbox.com
