lkml.org 
[lkml]   [2021]   [Oct]   [22]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH 1/2] sched/fair: Couple wakee flips with heavy wakers
From
Date
On Thu, 2021-10-21 at 15:56 +0100, Mel Gorman wrote:
>
> From additional tests on various servers, the impact is machine dependant
> but generally this patch improves the situation.
>
> hackbench-process-pipes
>                           5.15.0-rc3             5.15.0-rc3
>                              vanilla  sched-wakeeflips-v1r1
> Amean     1        0.3667 (   0.00%)      0.3890 (  -6.09%)
> Amean     4        0.5343 (   0.00%)      0.5217 (   2.37%)
> Amean     7        0.5300 (   0.00%)      0.5387 (  -1.64%)
> Amean     12       0.5737 (   0.00%)      0.5443 (   5.11%)
> Amean     21       0.6727 (   0.00%)      0.6487 (   3.57%)
> Amean     30       0.8583 (   0.00%)      0.8033 (   6.41%)
> Amean     48       1.3977 (   0.00%)      1.2400 *  11.28%*
> Amean     79       1.9790 (   0.00%)      1.8200 *   8.03%*
> Amean     110      2.8020 (   0.00%)      2.5820 *   7.85%*
> Amean     141      3.6683 (   0.00%)      3.2203 *  12.21%*
> Amean     172      4.6687 (   0.00%)      3.8200 *  18.18%*
> Amean     203      5.2183 (   0.00%)      4.3357 *  16.91%*
> Amean     234      6.1077 (   0.00%)      4.8047 *  21.33%*
> Amean     265      7.1313 (   0.00%)      5.1243 *  28.14%*
> Amean     296      7.7557 (   0.00%)      5.5940 *  27.87%*
>
> While different machines showed different results, in general
> there were much less CPU migrations of tasks

Patchlet helped hackbench? That's.. unexpected (at least by me).

> tbench4
>                            5.15.0-rc3             5.15.0-rc3
>                               vanilla  sched-wakeeflips-v1r1
> Hmean     1         824.05 (   0.00%)      802.56 *  -2.61%*
> Hmean     2        1578.49 (   0.00%)     1645.11 *   4.22%*
> Hmean     4        2959.08 (   0.00%)     2984.75 *   0.87%*
> Hmean     8        5080.09 (   0.00%)     5173.35 *   1.84%*
> Hmean     16       8276.02 (   0.00%)     9327.17 *  12.70%*
> Hmean     32      15501.61 (   0.00%)    15925.55 *   2.73%*
> Hmean     64      27313.67 (   0.00%)    24107.81 * -11.74%*
> Hmean     128     32928.19 (   0.00%)    36261.75 *  10.12%*
> Hmean     256     35434.73 (   0.00%)    38670.61 *   9.13%*
> Hmean     512     50098.34 (   0.00%)    53243.75 *   6.28%*
> Hmean     1024    69503.69 (   0.00%)    67425.26 *  -2.99%*
>
> Bit of a mixed bag but wins more than it loses.

Hm. If patchlet repeatably impacts buddy pairs one way or the other,
it should probably be tossed out the nearest window.


-Mike
\
 
 \ /
  Last update: 2021-10-22 12:27    [W:0.110 / U:0.548 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site