Re: [RFC PATCH 14/14] sched: implement usage tracking

[Peter Zijlstra]

On Wed, 2012-03-14 at 08:47 -0700, Paul Turner wrote:
> On Wed, Mar 14, 2012 at 8:01 AM, Peter Zijlstra <a.p.zijlstra@xxxxxxxxx> wrote:
> > On Tue, 2012-03-13 at 17:57 +0100, Vincent Guittot wrote:
> >> >  struct sched_avg {
> >> >        u64 runnable_avg_sum, runnable_avg_period;
> >> >        u64 last_runnable_update, decay_count;
> >> > +       u32 usage_avg_sum;
> >>
> >> Why usage_avg_sum is 32bits whereas runnable_avg_sum and
> >> runnable_avg_period are 64bits long ? You are doing the same
> >> computation on these 3 variables. Only the computation need to be done
> >> in 64bits but the result could be saved in 32bits ?
> >
> > Since you can never use more than 100% of cpu time, usage_avg_sum is
> > bound to 100% of the period, which (assuming your period < ~4s) should
> > thus still fit in the ~4s u32 provides.
> >
> > Runnable otoh is not bound by that and thus we cannot guarantee the
> > value stays within the ~4s value range.
> 
> Actually for runnable we can also make such a guarantee since:
> 
> runnable_sum <= \Sum 1024 * k^p --> 1024/(1-k) [geometric series, k<1]
> --> ~48k for our choice of k.
> 
> (We do however need 64-bits on any values that accumulate sums of
> loads, e.g. removed_load and *_load_sum.)

Only for the single entry, right? For the aggregated case the runnable
count can be nr_running times your limit, and IIRC (but my brain is
fuzzy since its been a while since I looked at this stuff) you use the
same structure in both cases.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/