Re: [PATCH v2 4/5] sched/pelt: Add a new runnable average signal

[Vincent Guittot]

On Tue, 18 Feb 2020 at 22:19, Valentin Schneider
<valentin.schneider@xxxxxxx> wrote:
>
> On 14/02/2020 15:27, Vincent Guittot wrote:
> > Now that runnable_load_avg has been removed, we can replace it by a new
> > signal that will highlight the runnable pressure on a cfs_rq. This signal
> > track the waiting time of tasks on rq and can help to better define the
> > state of rqs.
> >
> > At now, only util_avg is used to define the state of a rq:
> >   A rq with more that around 80% of utilization and more than 1 tasks is
> >   considered as overloaded.
> >
> > But the util_avg signal of a rq can become temporaly low after that a task
> > migrated onto another rq which can bias the classification of the rq.
> >
> > When tasks compete for the same rq, their runnable average signal will be
> > higher than util_avg as it will include the waiting time and we can use
> > this signal to better classify cfs_rqs.
> >
> > The new runnable_avg will track the runnable time of a task which simply
> > adds the waiting time to the running time. The runnable _avg of cfs_rq
> > will be the /Sum of se's runnable_avg and the runnable_avg of group entity
> > will follow the one of the rq similarly to util_avg.
> >
>
> I did a bit of playing around with tracepoints and it seems to be behaving
> fine. For instance, if I spawn 12 always runnable tasks (sysbench --test=cpu)
> on my Juno (6 CPUs), I get to a system-wide runnable value (\Sum cpu_runnable())
> of about 12K. I've only eyeballed them, but migration of the signal values
> seem fine too.
>
> I have a slight worry that the rq-wide runnable signal might be too easy to
> inflate, since we aggregate for *all* runnable tasks, and that may not play
> well with your group_is_overloaded() change (despite having the imbalance_pct
> on the "right" side).
>
> In any case I'll need to convince myself of it with some messing around, and
> this concerns patch 5 more than patch 4. So FWIW for this one:
>
> Tested-by: Valentin Schneider <valentin.schneider@xxxxxxx>
>
> I also have one (two) more nit(s) below.
>
> > Signed-off-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
> > ---
> > diff --git a/kernel/sched/pelt.c b/kernel/sched/pelt.c
> > @@ -227,14 +231,14 @@ ___update_load_sum(u64 now, struct sched_avg *sa,
> >        * Step 1: accumulate *_sum since last_update_time. If we haven't
> >        * crossed period boundaries, finish.
> >        */
> > -     if (!accumulate_sum(delta, sa, load, running))
> > +     if (!accumulate_sum(delta, sa, load, runnable, running))
> >               return 0;
> >
> >       return 1;
> >  }
> >
> >  static __always_inline void
> > -___update_load_avg(struct sched_avg *sa, unsigned long load)
> > +___update_load_avg(struct sched_avg *sa, unsigned long load, unsigned long runnable)
> >  {
> >       u32 divider = LOAD_AVG_MAX - 1024 + sa->period_contrib;
> >
> > @@ -242,6 +246,7 @@ ___update_load_avg(struct sched_avg *sa, unsigned long load)
> >        * Step 2: update *_avg.
> >        */
> >       sa->load_avg = div_u64(load * sa->load_sum, divider);
> > +     sa->runnable_avg =      div _u64(runnable * sa->runnable_sum, divider);
>                           ^^^^^^        ^^^^^^^^
>                             a)             b)
> a) That's a tab
>
> b) The value being passed is always 1, do we really need it to expose it as a
>    parameter?

In fact, I haven't been able to convince myself if it was better to
add the SCHED_CAPACITY_SCALE range in the _sum or only in the _avg.
That's the reason for this parameter to still being there.
On one side we do a shift at every PELT update and the
attach/detach/propagate are quite straight forward. On the other side
it is done only during attach/detach/propagate but it complexify the
thing. Having it in _sum doesn't seem to be a concern so I will keep
it there and remove the parameter