Re: [RFC patch 1/2] sched: dynamically adapt granularity with nr_running

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

* Peter Zijlstra (peterz@infradead.org) wrote:
> On Sun, 2010-09-12 at 11:06 +0200, Peter Zijlstra wrote:
> >  * is slice length, + is period length
> > 
> > Patch (sched_latency = 10, sched_min_gran = 10/3)
> > 
> > 
> > 30 |                             +
> >    |
> >    |
> >    |                          +
> >    |
> >    |
> >    |
> >    |
> >    |
> >    |
> > 20 |
> >    |
> >    |
> >    |
> >    |
> >    |
> >    |
> >    |
> >    |
> >    |
> > 10 |  *  +  +  +  +  +  +  +
> >    |
> >    |
> >    |
> >    |
> >    |     *
> >    |
> >    |        *  *              *  *  *  *  *  *
> >    |              *  *
> >    |                    *  *
> > 0  +---------------------------------------------------------
> >    0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 
> 
> 
> Vertical is time , horz is nr_running.
> 
> 
> Note how things fall of a cliff when nr_running goes from 8 to 9. Andrew
> usually kicks people in the teeth for phase change behaviour.

Please refer to my other email asking why your sched_latency = 10 rather than
what I see in sched_fair.c:

unsigned int sysctl_sched_latency = 6000000ULL;

I really might have missed something important there. Is this value recomputed
dynamically somewhere ?

Thanks,

Mathieu

-- 
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com