Re: [PATCH 0/8] sched/fair: misfit task load-balance tweaks

[Dietmar Eggemann <dietmar.eggemann@arm.com>]

On 03/02/2021 19:43, Valentin Schneider wrote:
> Hi Qais,
> 
> On 03/02/21 15:14, Qais Yousef wrote:
>> On 01/28/21 18:31, Valentin Schneider wrote:
>>> Hi folks,
>>>
>>> Here is this year's series of misfit changes. On the menu:
>>>
>>> o Patch 1 is an independent active balance cleanup
>>> o Patch 2 adds some more sched_asym_cpucapacity static branches
>>> o Patch 3 introduces yet another margin for capacity to capacity
>>>   comparisons
>>> o Patches 4-6 build on top of patch 3 and change capacity comparisons
>>>   throughout misfit load balancing  
>>> o Patches 7-8 fix some extra misfit issues I've been seeing on "real"
>>>   workloads.
>>>
>>> IMO the somewhat controversial bit is patch 3, because it attempts to solve
>>> margin issues by... Adding another margin. This does solve issues on
>>> existing platforms (e.g. Pixel4), but we'll be back to square one the day
>>> some "clever" folks spin a platform with two different CPU capacities less than
>>> 5% apart.
>>
>> One more margin is a cause of apprehension to me. But in this case I think it
>> is the appropriate thing to do now. I can't think of a scenario where this
>> could hurt.
>>
> 
> Thanks for having a look!
> 
>> Thanks
>>
>> --
>> Qais Yousef

How did you verify the benefit of these changes?

It's clear that you need a platform with capacity_orig diffs <20%
between CPU types (like Pixel4 - SD855 (4x261, 3x871, 1x1024) or QC's
RB5 platform - SD865 (4x284, 3x871, 1*1024)) but which
benchmark/testcase did you use?