lkml.org 
[lkml]   [2019]   [Oct]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [RFC PATCH v3 00/16] Core scheduling v3
From
Date
On Tue, 2019-10-29 at 10:11 +0100, Dario Faggioli wrote:
> On Sun, 2019-09-15 at 22:14 +0800, Aaron Lu wrote:
> > I'm using the following branch as base which is v5.1.5 based:
> > https://github.com/digitalocean/linux-coresched coresched-v3-
> > v5.1.5-
> > test
> >
> > And I have pushed Tim's branch to:
> > https://github.com/aaronlu/linux coresched-v3-v5.1.5-test-tim
> >
> > Mine:
> > https://github.com/aaronlu/linux coresched-v3-v5.1.5-test-
> > core_vruntime
> >
> Hello,
>
> As anticipated, I've been trying to follow the development of this
> feature and, in the meantime, I have done some benchmarks.
>
> I actually have a lot of data (and am planning for more), so I am
> sending a few emails, each one with a subset of the numbers in it,
> instead than just one which would be beyond giant! :-)
>
SYSBENCH
========

http://xenbits.xen.org/people/dariof/benchmarks/results/linux/core-sched/mmtests/boxes/wayrath/coresched-email-6_sysbench.txt

v v BM BM BM BM BM BM
BM-HT BM-noHT HT noHT csc-HT csc_stallfix-HT csc_tim-HT csc_vruntime-HT
Hmean 1 235.81 ( 0.00%) 221.49 ( -6.07%) 245.28 ( 4.01%) 230.53 ( -2.24%) 241.40 ( 2.37%) 225.00 ( -4.58%) 225.50 ( -4.37%) 202.38 ( -14.18%)
Hmean 4 273.77 ( 0.00%) 290.01 ( 5.93%) 292.47 ( 6.83%) 261.76 ( -4.39%) 287.58 ( 5.04%) 281.30 ( 2.75%) 274.21 ( 0.16%) 271.91 ( -0.68%)
Hmean 7 346.60 ( 0.00%) 315.58 ( -8.95%) 345.38 ( -0.35%) 349.29 ( 0.78%) 363.76 ( 4.95%) 349.09 ( 0.72%) 355.69 ( 2.62%) 336.69 ( -2.86%)
Hmean 8 343.17 ( 0.00%) 353.73 ( 3.08%) 409.04 ( 19.19%) 411.31 ( 19.86%) 406.77 ( 18.53%) 306.33 ( -10.74%) 393.70 ( 14.72%) 342.73 ( -0.13%)
Stddev 1 44.93 ( 0.00%) 50.07 ( -11.44%) 25.05 ( 44.24%) 39.22 ( 12.71%) 26.27 ( 41.54%) 42.77 ( 4.81%) 43.63 ( 2.90%) 62.88 ( -39.95%)
Stddev 4 16.03 ( 0.00%) 23.37 ( -45.76%) 23.77 ( -48.25%) 22.40 ( -39.69%) 18.63 ( -16.19%) 14.37 ( 10.35%) 9.34 ( 41.72%) 25.21 ( -57.23%)
Stddev 7 22.88 ( 0.00%) 37.54 ( -64.07%) 26.57 ( -16.16%) 38.50 ( -68.26%) 59.14 (-158.51%) 26.73 ( -16.83%) 24.58 ( -7.43%) 32.93 ( -43.94%)
Stddev 8 36.74 ( 0.00%) 36.60 ( 0.39%) 102.82 (-179.86%) 93.56 (-154.65%) 77.33 (-110.47%) 36.16 ( 1.58%) 44.27 ( -20.50%) 35.15 ( 4.33%)
v v VM VM VM VM VM VM
VM-HT VM-noHT HT noHT csc-HT csc_stallfix-HT csc_tim-HT csc_vruntime-HT
Hmean 1 215.16 ( 0.00%) 225.80 * 4.95%* 205.74 ( -4.38%) 200.61 ( -6.76%) 169.70 ( -21.13%) 168.84 ( -21.53%) 157.27 ( -26.91%) 162.94 ( -24.27%)
Hmean 4 163.44 ( 0.00%) 189.82 * 16.14%* 164.54 ( 0.67%) 148.47 ( -9.16%) 40.62 * -75.15%* 53.14 * -67.49%* 129.51 * -20.76%* 158.99 ( -2.72%)
Hmean 7 162.74 ( 0.00%) 185.79 ( 14.17%) 211.79 ( 30.14%) 186.92 ( 14.86%) 28.02 * -82.78%* 34.32 * -78.91%* 130.01 ( -20.11%) 145.47 ( -10.61%)
Hmean 8 240.19 ( 0.00%) 192.24 ( -19.96%) 192.87 * -19.70%* 194.01 ( -19.23%) 16.92 * -92.95%* 30.55 * -87.28%* 150.51 * -37.34%* 147.67 * -38.52%*
Stddev 1 1.80 ( 0.00%) 4.14 (-129.80%) 6.04 (-234.90%) 24.54 (-1261.18%) 61.54 (-3313.19%) 62.92 (-3389.48%) 55.91 (-3000.67%) 58.29 (-3132.75%)
Stddev 4 6.33 ( 0.00%) 14.22 (-124.51%) 7.04 ( -11.07%) 13.77 (-117.30%) 13.23 (-108.83%) 5.07 ( 19.92%) 15.73 (-148.38%) 14.04 (-121.61%)
Stddev 7 24.70 ( 0.00%) 37.50 ( -51.78%) 35.59 ( -44.07%) 29.96 ( -21.26%) 20.07 ( 18.77%) 21.06 ( 14.74%) 24.97 ( -1.07%) 30.52 ( -23.52%)
Stddev 8 23.23 ( 0.00%) 41.13 ( -77.03%) 27.30 ( -17.49%) 41.67 ( -79.35%) 3.75 ( 83.88%) 12.62 ( 45.68%) 55.87 (-140.49%) 37.12 ( -59.76%)
v v VM VM VM VM VM VM
VM-v4-HT VM-v4-noHT v4-HT v4-noHT v4-csc-HT v4-csc_stallfix-HT v4-csc_tim-HT v4-csc_vruntime-HT
Hmean 1 216.12 ( 0.00%) 310.43 ( 43.63%) 168.80 ( -21.90%) 196.04 ( -9.29%) 168.27 ( -22.14%) 188.53 ( -12.77%) 176.57 ( -18.30%) 180.12 ( -16.66%)
Hmean 3 161.91 ( 0.00%) 160.80 ( -0.69%) 160.33 ( -0.97%) 175.36 ( 8.31%) 51.27 * -68.34%* 52.18 * -67.77%* 137.95 ( -14.80%) 166.12 ( 2.60%)
Hmean 4 156.44 ( 0.00%) 196.25 * 25.45%* 165.19 ( 5.60%) 199.78 * 27.71%* 50.67 * -67.61%* 40.42 * -74.16%* 175.03 ( 11.88%) 172.66 * 10.37%*
Stddev 1 4.67 ( 0.00%) 100.18 (-2043.42%) 50.61 (-982.87%) 141.33 (-2923.68%) 69.08 (-1377.96%) 70.09 (-1399.51%) 47.32 (-912.42%) 42.67 (-813.02%)
Stddev 3 12.62 ( 0.00%) 8.76 ( 30.60%) 13.18 ( -4.38%) 9.42 ( 25.41%) 3.57 ( 71.72%) 28.47 (-125.51%) 21.08 ( -67.00%) 19.37 ( -53.48%)
Stddev 4 8.52 ( 0.00%) 8.54 ( -0.17%) 3.09 ( 63.70%) 13.84 ( -62.42%) 24.33 (-185.50%) 9.30 ( -9.10%) 16.60 ( -94.73%) 9.64 ( -13.07%)
v v VMx2 VMx2 VMx2 VMx2 VMx2 VMx2
VMx2-HT VMx2-noHT HT noHT csc-HT csc_stallfix-HT csc_tim-HT csc_vruntime-HT
Hmean 1 168.87 ( 0.00%) 154.79 ( -8.34%) 190.28 ( 12.68%) 151.18 ( -10.48%) 136.73 ( -19.03%) 36.63 * -78.31%* 83.21 ( -50.73%) 124.42 ( -26.32%)
Hmean 4 163.65 ( 0.00%) 87.90 * -46.29%* 119.37 * -27.06%* 87.94 * -46.26%* 26.96 * -83.53%* 24.08 * -85.29%* 54.15 * -66.91%* 63.80 * -61.01%*
Hmean 7 181.60 ( 0.00%) 89.10 * -50.93%* 148.16 ( -18.41%) 75.71 * -58.31%* 16.98 * -90.65%* 23.92 * -86.83%* 57.28 * -68.46%* 66.10 * -63.60%*
Hmean 8 198.98 ( 0.00%) 94.24 * -52.64%* 141.96 ( -28.65%) 96.62 * -51.44%* 23.22 * -88.33%* 29.24 * -85.30%* 80.10 * -59.74%* 80.36 * -59.61%*
Stddev 1 61.59 ( 0.00%) 44.71 ( 27.41%) 52.14 ( 15.33%) 44.61 ( 27.56%) 90.12 ( -46.32%) 42.53 ( 30.94%) 38.32 ( 37.79%) 43.03 ( 30.12%)
Stddev 4 8.65 ( 0.00%) 21.74 (-151.41%) 21.18 (-144.98%) 22.51 (-160.27%) 19.72 (-128.07%) 4.38 ( 49.38%) 2.68 ( 68.95%) 14.40 ( -66.55%)
Stddev 7 17.94 ( 0.00%) 15.14 ( 15.62%) 29.94 ( -66.88%) 17.30 ( 3.54%) 26.23 ( -46.20%) 5.17 ( 71.17%) 15.98 ( 10.95%) 18.43 ( -2.72%)
Stddev 8 38.45 ( 0.00%) 19.68 ( 48.82%) 44.14 ( -14.78%) 28.84 ( 25.00%) 10.64 ( 72.33%) 11.65 ( 69.71%) 22.63 ( 41.15%) 16.00 ( 58.39%)

Core scheduling does not seem to be able to handle sysbench well, yet.
In this case, things are not to bad on baremtal (and the best
performing coresched variant is again the one with Tim's patches).

But things go bad when running the banchmark in VMs, where core
scheduling almost always loses against no HyperThreading, even in the
overcommitted case. For virt. cases, it's also not straightforward to
tell which set of patches are best, as some runs are in favours of
Tim's, some others of vruntime's.

--
Dario Faggioli, Ph.D
http://about.me/dario.faggioli
Virtualization Software Engineer
SUSE Labs, SUSE https://www.suse.com/
-------------------------------------------------------------------
<<This happens because _I_ choose it to happen!>> (Raistlin Majere)

[unhandled content-type:application/pgp-signature]
\
 
 \ /
  Last update: 2019-10-29 10:19    [W:1.569 / U:0.068 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site