Re: [PATCH] sched/fair: update_pick_idlest() Select group with lowest group_util when idle_cpus are equal

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

On 17/06/2020 16:52, Peter Puhov wrote:
> On Wed, 17 Jun 2020 at 06:50, Valentin Schneider
> <valentin.schneider@arm.com> wrote:
>>
>>
>> On 16/06/20 17:48, peter.puhov@linaro.org wrote:
>>> From: Peter Puhov <peter.puhov@linaro.org>
>>> We tested this patch with following benchmarks:
>>>   perf bench -f simple sched pipe -l 4000000
>>>   perf bench -f simple sched messaging -l 30000
>>>   perf bench -f simple  mem memset -s 3GB -l 15 -f default
>>>   perf bench -f simple futex wake -s -t 640 -w 1
>>>   sysbench cpu --threads=8 --cpu-max-prime=10000 run
>>>   sysbench memory --memory-access-mode=rnd --threads=8 run
>>>   sysbench threads --threads=8 run
>>>   sysbench mutex --mutex-num=1 --threads=8 run
>>>   hackbench --loops 20000
>>>   hackbench --pipe --threads --loops 20000
>>>   hackbench --pipe --threads --loops 20000 --datasize 4096
>>>
>>> and found some performance improvements in:
>>>   sysbench threads
>>>   sysbench mutex
>>>   perf bench futex wake
>>> and no regressions in others.
>>>
>>
>> One nitpick for the results of those: condensing them in a table form would
>> make them more reader-friendly. Perhaps something like:
>>
>> | Benchmark        | Metric   | Lower is better? | BASELINE | SERIES | DELTA |
>> |------------------+----------+------------------+----------+--------+-------|
>> | Sysbench threads | # events | No               |    45526 |  56567 |  +24% |
>> | Sysbench mutex   | ...      |                  |          |        |       |
>>
>> If you want to include more stats for each benchmark, you could have one table
>> per (e.g. see [1]) - it'd still be a more readable form (or so I believe).

Wouldn't Unix Bench's 'execl' and 'spawn' be the ultimate test cases
for those kind of changes?

I only see minor improvements with tip/sched/core as base on hikey620
(Arm64 octa-core).

				base		w/ patch
./Run spawn -c 8 -i 10		 633.6		 635.1

./Run execl -c 8 -i 10		1187.5		1190.7	


At the end of find_idlest_group(), when comparing local and idlest, it
is explicitly mentioned that number of idle_cpus is used instead of
utilization.
The comparision between potential idle groups and local & idlest group
should probably follow the same rules.

I haven't tested it with 

https://lkml.kernel.org/r/20200624154422.29166-1-vincent.guittot@linaro.org

which might have an influence here too.