http://defect.opensolaris.org/bz/show_bug.cgi?id=4015





--- Comment #7 from Eric Saxe <eric.saxe at sun.com>  2008-10-22 10:11:20 ---
(In reply to comment #6)
> >Yes. I think we want to try and avoid dispatching threads to idle domains 
> >that
> >are sleeping deeply, and instead utilize a "shallower" idle domain, or one
> >that's not idle at all. I would worry that there would be issues with
> >re-leveling the hierarchy in this way (no sorting by # of CPUs). I had 
> >instead
> >thought of explicitly looking at the "idle depth" when searchign for an idle
> >CPU in a candidate PG group when forced to grow utilization to other 
> >domains...
> 
> I think it probably can be implemented passively? currently cmt pg's capacity
> is
> defined to be the number of active logical CPUs in the group. I think this
> number 
> is too small to help much for deep idle. Increasing the capacity 
> automatically 
> group the utilization on one idle domain and help CPUs in another domain 
> sleep 
> deeper.

I agree, I really like this approach. Modifying the capacities to tune the
aggressiveness of the coalescence policy is a good way to go.

-- 
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.

Reply via email to