Re: [PATCH v4] sched/fair: unlink misfit task from cpu overutilized

[Vincent Guittot]

On Mon, 23 Jan 2023 at 13:10, Kajetan Puchalski
<kajetan.puchalski@xxxxxxx> wrote:
>
> > By taking into account uclamp_min, the 1:1 relation between task misfit
> > and cpu overutilized is no more true as a task with a small util_avg may
> > not fit a high capacity cpu because of uclamp_min constraint.
> >
> > Add a new state in util_fits_cpu() to reflect the case that task would fit
> > a CPU except for the uclamp_min hint which is a performance requirement.
> >
> > Use -1 to reflect that a CPU doesn't fit only because of uclamp_min so we
> > can use this new value to take additional action to select the best CPU
> > that doesn't match uclamp_min hint.
> >
> > Signed-off-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
> > ---
> >
> > Change since v3:
> > - Keep current condition for uclamp_max_fits in util_fits_cpu()
> > - Update some comments
>
> That one condition change from v3 did fix the overutilization issues so
> good news on that front :)
>
> 1. GB5
>
> +-----------------+-------------------------+--------+-----------+
> |     metric      |         kernel          | value  | perc_diff |
> +-----------------+-------------------------+--------+-----------+
> | multicore_score |        baseline         | 2765.4 |   0.0%    |
> | multicore_score |      baseline_ufc       | 2704.3 |  -2.21%   | <-- current mainline regression
> | multicore_score |     ufc_patched_v4      | 2839.8 |   2.69%   | <-- new score improvement
> +-----------------+-------------------------+--------+-----------+
>
> +--------------+--------+-------------------------+--------+-----------+
> |  chan_name   | metric |         kernel          | value  | perc_diff |
> +--------------+--------+-------------------------+--------+-----------+
> | total_power  | gmean  |        baseline         | 2664.0 |   0.0%    |
> | total_power  | gmean  |      baseline_ufc       | 2621.5 |   -1.6%   |
> | total_power  | gmean  |     ufc_patched_v4      | 2729.0 |   2.44%   |
> +--------------+--------+-------------------------+--------+-----------+
>
> 2. Jankbench
>
> +--------+---------------+------------------------------+-------+-----------+
> | metric |   variable    |            kernel            | value | perc_diff |
> +--------+---------------+------------------------------+-------+-----------+
> | gmean  | mean_duration |        baseline_60hz         | 14.6  |   0.0%    |
> | gmean  | mean_duration |      baseline_ufc_60hz       | 15.2  |   3.83%   |
> | gmean  | mean_duration |     ufc_patched_v4_60hz      | 14.0  |  -3.98%   |
> +--------+---------------+------------------------------+-------+-----------+
>
> +--------+-----------+------------------------------+-------+-----------+
> | metric | variable  |            kernel            | value | perc_diff |
> +--------+-----------+------------------------------+-------+-----------+
> | gmean  | jank_perc |        baseline_60hz         |  1.9  |   0.0%    |
> | gmean  | jank_perc |      baseline_ufc_60hz       |  2.2  |  15.39%   |
> | gmean  | jank_perc |     ufc_patched_v4_60hz      |  1.8  |  -5.67%   |
> +--------+-----------+------------------------------+-------+-----------+
>
> +--------------+--------+------------------------------+-------+-----------+
> |  chan_name   | metric |            kernel            | value | perc_diff |
> +--------------+--------+------------------------------+-------+-----------+
> | total_power  | gmean  |        baseline_60hz         | 135.9 |   0.0%    |
> | total_power  | gmean  |      baseline_ufc_60hz       | 155.7 |  14.61%   | <-- current mainline regression
> | total_power  | gmean  |     ufc_patched_v4_60hz      | 131.4 |  -3.26%   | <-- new power saving
> +--------------+--------+------------------------------+-------+-----------+
>
> All in all this comes out better on every metric than the previous
> baseline and way better than current mainline. At least from an Android
> perspective as far as the impacts go I'd say it's probably fine to go
> ahead and apply this.

Thanks for your tests results

>
> Feel free to add this if you'd like:
> Tested-by: Kajetan Puchalski <kajetan.puchalski@xxxxxxx>
>
> > --
> > 2.34.1
> >
> >