Re: [PATCH V4] intel_pstate: Use avg_pstate instead of current_pstate

[Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>]

Hi Doug,

On Fri, 2016-04-22 at 08:18 -0700, Doug Smythies wrote:
> Srinivas,
> 
> Recall a couple of months ago, on the "Increase hold-off time before
> busyness is scaled"
> thread, Stephane suggested we try applying this method on the
> get_target_pstate_use_performance
> branch of the intel_pstate driver, as opposed to just on the
> get_target_pstate_use_cpu_load branch.
> 
> It didn't make any difference with respect to the hold-off time
> issue.
> However, I did spend considerable time testing it in other scenarios.
> It does somewhat temper the occasional tendency to suddenly have a
> ridiculously high
> scaled busy number with virtually no load (the same issue from the
> "[intel-pstate driver regression] processor frequency very high even
> if in idle" thread,
> that continued in https://bugzilla.kernel.org/show_bug.cgi?id=115771 
> )
> I didn't find any significant regression, and did observe some small
> energy savings
> in some scenarios (admittedly, small enough to have possibly been
> simply test to test
> variations, and I didn't do enough tests to extract a definite
> trend).
> 
> I am suggesting to consider extending the patch to
> get_target_pstate_use_performance also.
> 
Many core platforms have per core P states. So here the assumption that
we get a boost of P State on one core because some activity on other
core will not hold true.

We are experimenting with algorithm to improve core performance
(similar approach as yours + IO boost), hopefully we can publish soon.

Thanks,
Srinivas


> ... Doug
> 
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-pm"
> in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html