[tesla-dev] [Bug 4616] cmt_balance schedules thread on any idle CPU instead of shallow idle CPUs

[bugzilla-daemon at defect.opensolaris.org]

http://defect.opensolaris.org/bz/show_bug.cgi?id=4616





--- Comment #5 from Bill Holler <bill.holler at sun.com>  2008-11-25 21:18:34 ---
cmt_select_idle_cpu was set with mdb to either the default selection policy
and the cstate-aware selection policy (cmt_select_idle_cpu_dflt
and cpu_select_idle_cpu).  With c-states disabled performance is the same
for both policies.  This shows the new function is not noticeably slower
than the old function.

With c-states enabled switching from the non-cstate aware selection function
to the c-state aware function causes system time in C2 to drop in half and
system time in C3 to double with a fork workload.  Very interesting.
The slowdown is happening because this new idle cpu selection algorithm
is much better at allowing C3 cpus to stay in C3 long enough to actually
enter a power saving state with associated increased wakeup latency.

-- 
Configure bugmail: http://defect.opensolaris.org/bz/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.