Re: [patch] Re: scheduler oddity [bug?]

[Mike Galbraith <efault@gmx.de>]

On Mon, 2009-03-09 at 14:27 +0100, Peter Zijlstra wrote:
> On Mon, 2009-03-09 at 14:16 +0100, Mike Galbraith wrote:
> > On Mon, 2009-03-09 at 12:04 +0100, Peter Zijlstra wrote:
> > 
> > > OK, talked a bit with Ingo, the reason you're doing is that avg_overlap
> > > can easily grow stale.. I can see that happen indeed.
> > > 
> > > So the 'perfect' thing would be a task-runtime decay, barring that the
> > > preemption thing seems a sane enough hart-beat of a task.
> > > 
> > > How does the below look to you?
> > 
> > Other than the fact that the test for sync reject is currently
> > avg_overlap > sysctl_sched_migration_cost, looks fine to me.  Having it
> > capped at the boundary is probably the better way to go.
> 
> Ah, yes, and looking at update_avg() we'll also discard the lower 3
> bits, so we'll never actually reach.
> 
> So I guess it should read something like:
> 
>   update_avg(&prev->se.avg_overlap, 2*sysctl_sched_migration_cost);
> 
> or somesuch.
> 
> Does it actually solve the reported problem? I've only thought about the
> issue so far :-)

 5977 root      20   0  3672  440  352 R  100  0.0   0:28.53 2 pipetest
 5978 root      20   0  3668  180   96 S   29  0.0   0:08.27 0 pipetest

Yup, works for me.  Ship it :)

	-Mike