Messages in this thread | | | From | "Rafael J. Wysocki" <> | Date | Mon, 21 Nov 2016 21:53:44 +0100 | Subject | Re: [PATCH] cpufreq: schedutil: add up/down frequency transition rate limits |
| |
On Mon, Nov 21, 2016 at 5:46 PM, Peter Zijlstra <peterz@infradead.org> wrote: > On Mon, Nov 21, 2016 at 04:24:24PM +0000, Patrick Bellasi wrote: >> On 21-Nov 16:26, Peter Zijlstra wrote: > >> > In any case, worth trying, see what happens. >> >> Are you saying that you would like to see the code which implements a >> more generic version of the peak_util "filter" on top of PELT? > > Not sure about peak_util, I was more thinking of an IIR/PID filter, as > per the email thread referenced below. Doesn't make sense to hide that > in intel_pstate if it appears to be universally useful etc.. > >> IMO it could be a good exercise now that we agree we want to improve >> PELT without replacing it. > > I think it would make sense to keep it inside sched_cpufreq for now. > >> > > For example, a task running 30 [ms] every 100 [ms] is a ~300 util_avg >> > > task. With PELT, we get a signal which range between [120,550] with an >> > > average of ~300 which is instead completely ignored. By capping the >> > > decay we will get: >> > > >> > > decay_cap [ms] range average >> > > 0 120:550 300 >> > > 64 140:560 310 >> > > 32 320:660 430 >> > > >> > > which means that still the raw PELT signal is wobbling and never >> > > provides a consistent response to drive decisions. >> > > >> > > Thus, a "predictor" should be something which sample information from >> > > PELT to provide a more consistent view, a sort of of low-pass filter >> > > on top of the "dynamic metric" which is PELT. >> > > >> > > Should not such a "predictor" help on solving some of the issues >> > > related to PELT slow ramp-up or fast ramp-down? >> > >> > I think intel_pstate recently added a local PID filter, I asked at the >> > time if something like that should live in generic code, looks like >> > maybe it should. >> >> That PID filter is not "just" a software implementation of the ACPI's >> Collaborative Processor Performance Control (CPPC) when HWP hardware >> is not provided by a certain processor? > > I think it was this thread: > > http://lkml.kernel.org/r/1572483.RZjvRFdxPx@vostro.rjw.lan > > It never really made sense such a filter should live in individual > drivers.
We don't use the IIR filter in intel_pstate after all.
We evaluated it, but it affected performance too much to be useful for us.
That said in the "proportional" version of the intel_pstate's P-state selection algorithm (without PID) we ramp up faster than we reduce the P-state, but the approach used in there depends on using the feedback registers.
And, of course, that's only used if HWP is not active.
Thanks, Rafael
| |