[PATCH v2 08/11] sched: get CPU's activity statistic

[Morten Rasmussen]

On Wed, Jun 04, 2014 at 02:09:18PM +0100, Morten Rasmussen wrote:
> On Wed, Jun 04, 2014 at 12:52:46PM +0100, Vincent Guittot wrote:
> > On 4 June 2014 13:23, Morten Rasmussen <morten.rasmussen at arm.com> wrote:
> > > I get:
> > >
> > > A: 15/40 ms = 37.5%
> > > B: 20/40 ms = 50%
> > >
> > > Schedule:
> > >
> > >    | 5 ms | 5 ms | 5 ms | 5 ms | 5 ms | 5 ms | 5 ms | 5 ms | 5 ms |
> > > A:   run     rq     run  ----------- sleeping -------------  run
> > > B:   rq      run    rq    run   ---- sleeping -------------  rq
> > >
> > >> and CPU runnable will be 60% too
> > >
> > > rq->avg.runnable_avg_sum should be 50%. You have two tasks running for
> > > 20 ms every 40 ms.
> > >
> > > Right?
> > 
> > ok, i see the misunderstood.
> > it's depends of what we mean by runnable. You take the % of time
> > whereas i take the runnable_avg_sum/period
> 
> Right. There is a difference.
> 
> > 
> > so A is on_rq 15/40 ms = 37.5% of the time which gives a
> > runnable_avg_sum/runnable_avg_period of 47%
> > B is on_rq  20/40 ms = 50% of the time which gives a
> > runnable_avg_sum/runnable_avg_period of 60%
> > and CPU has a task on its rq 20/40ms = 50% of the time which gives a
> > runnable_avg_sum/runnable_avg_period of 60%
> 
> Yes, that seems about right.

If my calculations are right, cfs.runnable_load_avg would peak 15 ms
into the period at about 104%. After 20 ms A has decayed more than B has
gained so the sum is slightly lower.