On 02/20/2013 06:49 PM, Ingo Molnar wrote:
[snip]
>
> The changes look clean and reasoable, any ideas exactly *why* it
> speeds up?
>
> I.e. are there one or two key changes in the before/after logic
> and scheduling patterns that you can identify as causing the
> speedup?
Hi, Ingo
Thanks for your reply, please let me point out the key changes here
(forgive me for haven't wrote a good description in cover).
The performance improvement from this patch set is:
1. delay the invoke on wake_affine().
2. save the circle to gain proper sd.
The second point is obviously, and will benefit a lot when the sd
topology is deep (NUMA is suppose to make it deeper on large system).
So in my testing on a 12 cpu box, actually most of the benefit comes
from the first point, and please let me introduce it in detail.
The old logical when locate affine_sd is:
if prev_cpu != curr_cpu
if wake_affine()
prev_cpu = curr_cpu
new_cpu = select_idle_sibling(prev_cpu)
return new_cpu
The new logical is same to the old one if prev_cpu == curr_cpu, so let's
simplify the old logical like:
if wake_affine()
new_cpu = select_idle_sibling(curr_cpu)
else
new_cpu = select_idle_sibling(prev_cpu)
return new_cpu
Actually that doesn't make sense.
I think wake_affine() is trying to check whether move a task from
prev_cpu to curr_cpu will break the balance in affine_sd or not, but why
won't break balance means curr_cpu is better than prev_cpu for searching
the idle cpu?
So the new logical in this patch set is:
new_cpu = select_idle_sibling(prev_cpu)
if idle_cpu(new_cpu)
return new_cpu
new_cpu = select_idle_sibling(curr_cpu)
if idle_cpu(new_cpu) {
if wake_affine()
return new_cpu
}
return prev_cpu
And now, unless we are really going to move load from prev_cpu to
curr_cpu, we won't use wake_affine() any more.
So we avoid wake_affine() when system load is low or high, for middle
load, the worst cases is when failed to locate idle cpu in prev_cpu
topology but succeed to locate one in curr_cpu's, but that's rarely
happen and the benchmark results proved that point.
Some comparison below:
1. system load is low
old logical cost:
wake_affine()
select_idle_sibling()
new logical cost:
select_idle_sibling()
2. system load is high
old logical cost:
wake_affine()
select_idle_sibling()
new logical cost:
select_idle_sibling()
select_idle_sibling()
3. system load is middle
don't know
1 save the cost of wake_affine(), 3 could be proved by benchmark that no
regression at least.
For 2, it's the comparison between wake_affine() and
select_idle_sibling(), since the system load is high, wake_affine() cost
far more than select_idle_sibling(), and we saved many according to the
benchmark results.
>
> Such changes also typically have a chance to cause regressions
> in other workloads - when that happens we need this kind of
> information to be able to enact plan-B.
The benefit comes from avoiding unnecessary works, and the patch set is
suppose to only reduce the cost of key function with least logical
changing, I could not promise it benefit all the workloads, but till
now, I've not found regression.
Regards,
Michael Wang
>
> Thanks,
>
> Ingo
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
