Re: [PATCH 3/3] sched, fair: Stop searching for tasks in newidle balance if there are runnable tasks

[Jason Low <jason.low2@hp.com>]

On Thu, 2014-04-24 at 18:52 +0200, Peter Zijlstra wrote:
> On Thu, Apr 24, 2014 at 09:43:09AM -0700, Jason Low wrote:
> > If the below patch is what you were referring to, I believe this
> > can help too. This was also something that I was testing out before
> > we went with those patches which compares avg_idle with idle balance
> > cost. I recall seeing somewhere around a +7% performance improvement
> > in at least least 1 of the AIM7 workloads. I can do some more testing
> > with this.
> 
> Yes, exactly that.
> 
> I can't remember the details, but I suspect we feared the less agressive
> idle balance due to that patch (it will only pull a single task, instead
> of multiple) would cause more idle_balance invocations and thereby
> decrease throughput.
> 
> So I suppose something with _many_ bursty threads which leads to severe
> inequalities would be the workload to trigger that.
> 
> Not sure we've ever seen that.. maybe Mike remembers, he seems to have a
> head for such details.

Okay, so running the AIM7 fserver workload, I didn't see any noticeable
performance gains with having move_tasks() pull at most one task. The
+7% performance improvement that I saw was without the idle balance cost
patches. I think that with those idle balance cost patches, there aren't
as much benefits with this patch, and allowing more than 1 task to be
pulled in move_task(), like we have now, may be the best option.