Re: [PATCH v2 00/13] sched: Clean-ups and asymmetric cpu capacity support

From: Morten Rasmussen
Date: Thu Jun 30 2016 - 03:52:42 EST


On Tue, Jun 28, 2016 at 06:20:27PM +0800, Koan-Sin Tan wrote:
> Hi,
>
> I tested these patches with patches from your preview branch [1] and some
> platform specific patches on MediaTek MT8173 EVB (integrated branch at [2])
>
> > Test 0:
> > for i in `seq 1 10`; \
> > do sysbench --test=cpu --max-time=3 --num-threads=1 run; \
> > done \
> > | awk '{if ($4=="events:") {print $5; sum +=$5; runs +=1}} \
> > END {print "Average events: " sum/runs}'
> > Target: ARM TC2 (2xA15+3xA7)
> >
> > (Higher is better)
> > tip: Average events: 146.9
> > patch: Average events: 217.9
>
> Target: MediaTek MT8173 EVB (2xA53+2xA72)
>
> w/o capacity patches:
> 274.7 (actually, it varies a lot between 194 and 329

The variation you mention is consistent with our observations on both
TC2 and Juno. Most times you get either big or little performance, and
occasionally a number in between.

> w/ capacity patches:
> 328.7
>
> > Test 1:
> > perf stat --null --repeat 10 -- \
> > perf bench sched messaging -g 50 -l 5000
> >
> > Target: Intel IVB-EP (2*10*2)
> >
> > tip: 4.861970420 seconds time elapsed ( +- 1.39% )
> > patch: 4.886204224 seconds time elapsed ( +- 0.75% )
> >
> > Target: ARM TC2 A7-only (3xA7) (-l 1000)
> >
> > tip: 61.485682596 seconds time elapsed ( +- 0.07% )
> > patch: 62.667950130 seconds time elapsed ( +- 0.36% )
> >
>
> Target: MediaTek MT8173 EVB A53-only (2xA7) (-l 1000)
>
> w/o capacity patches:
> 72.663181227 seconds time elapsed ( +- 1.62% )
> 71.502263666 seconds time elapsed ( +- 0.45% )
> 73.707466212 seconds time elapsed ( +- 0.99% )
> 73.623523997 seconds time elapsed ( +- 1.85% )
> 74.150137066 seconds time elapsed ( +- 1.41% )
>
> w/ capacity patches:
> 70.360055056 seconds time elapsed ( +- 1.58% )
> 69.884793481 seconds time elapsed ( +- 0.78% )
> 73.395917166 seconds time elapsed ( +- 1.86% )
> 70.082440427 seconds time elapsed ( +- 1.66% )
> 69.365401758 seconds time elapsed ( +- 0.80% )
>
> Tested-by: Koan-Sin Tan <freedom.tan@xxxxxxxxxxxx>

Thanks for testing.

Morten