diff mbox series

sched/eevdf: Toggle eligibility through sched_feat

Message ID	20231013030213.2472697-1-youssefesmat@chromium.org
State	New
Headers	Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:7 as permitted sender) client-ip=2620:137:e000::3:7; From: Youssef Esmat <youssefesmat@chromium.org> To: LKML <linux-kernel@vger.kernel.org> Cc: peterz@infradead.org, bsegall@google.com, mingo@kernel.org, vincent.guittot@linaro.org, juri.lelli@redhat.com, dietmar.eggemann@arm.com, rostedt@goodmis.org, mgorman@suse.de, bristot@redhat.com, corbet@lwn.net, qyousef@layalina.io, chris.hyser@oracle.com, patrick.bellasi@matbug.net, pjt@google.com, pavel@ucw.cz, qperret@google.com, tim.c.chen@linux.intel.com, joshdon@google.com, timj@gnu.org, kprateek.nayak@amd.com, yu.c.chen@intel.com, joel@joelfernandes.org, efault@gmx.de, tglx@linutronix.de, wuyun.abel@bytedance.com, Youssef Esmat <youssefesmat@chromium.org> Subject: [PATCH] sched/eevdf: Toggle eligibility through sched_feat Date: Thu, 12 Oct 2023 22:02:13 -0500 Message-ID: <20231013030213.2472697-1-youssefesmat@chromium.org> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Precedence: bulk
Series	sched/eevdf: Toggle eligibility through sched_feat \| sched/eevdf: Toggle eligibility through sched_feat

Commit Message

Youssef Esmat Oct. 13, 2023, 3:02 a.m. UTC

  Interactive workloads see performance gains by disabling eligibility
checks (EEVDF->EVDF). Disabling the checks reduces the number of
context switches and delays less important work (higher deadlines/nice
values) in favor of more important work (lower deadlines/nice values).

That said, that can add large latencies for some work loads and as the
default is eligibility on, but allowing it to be turned off when
beneficial.

Signed-off-by: Youssef Esmat <youssefesmat@chromium.org>
Link: https://lore.kernel.org/lkml/CA+q576MS0-MV1Oy-eecvmYpvNT3tqxD8syzrpxQ-Zk310hvRbw@mail.gmail.com/
---
 kernel/sched/fair.c     | 3 +++
 kernel/sched/features.h | 1 +
 2 files changed, 4 insertions(+)

Comments

Mike Galbraith Oct. 13, 2023, 6:50 a.m. UTC | #1

On Thu, 2023-10-12 at 22:02 -0500, Youssef Esmat wrote:
> Interactive workloads see performance gains by disabling eligibility
> checks (EEVDF->EVDF). Disabling the checks reduces the number of
> context switches and delays less important work (higher deadlines/nice
> values) in favor of more important work (lower deadlines/nice values).

For some definitions of both interactive and benefit, both being a bit
too slippery to nail to the floor.

> That said, that can add large latencies for some work loads and as the
> default is eligibility on, but allowing it to be turned off when
> beneficial.

The only contra I would mention WRT the feature is that features having
no abi or thou shalt not regress protections makes them.. suboptimal
tuning knobs for userspace.

WRT patchlet, I tried same yesterday for grins (but spelled it "EVDF")
and fed it my favorite mixed load of youtube vs 8ms run 1ms sleep
massive_intr as a vaguely kbuild like competing "compute" component. It
didn't do anything lovely for the "interactive" bits, in fact did some
damage when letting the CPU do the rendering (something I have good but
irrelevant reasons to do at times;).  With or without RUN_TO_PARITY,
EVDF skewed distribution toward "compute".

Some numbers, these from my old i7-4790 box:

Executive summary
  Task                  |   Runtime ms  | Switches | Avg delay ms    | Max delay ms    | Sum delay ms     |
stock
  massive_intr:(9)      |1827981.908 ms |  9163472 | avg:   0.040 ms | max:  13.026 ms | sum:366012.231 ms |
  chrome:(8)            |  33240.153 ms |   231301 | avg:   0.167 ms | max:   9.843 ms | sum:38606.462 ms |
  TOTAL:                |2307670.845 ms | 25693098 |                 |       19.050 ms |   1110835.993 ms |
evdf
  massive_intr:(9)      |1882920.417 ms |   575899 | avg:   0.536 ms | max:  31.997 ms | sum:308728.289 ms |
  chrome:(12)           |  33353.359 ms |    79764 | avg:   1.082 ms | max:  22.964 ms | sum:86266.324 ms |
  TOTAL:                |2289006.437 ms |  5105092 |                 |       31.997 ms |   2748046.342 ms |
evdf+no_run_to_parity
  massive_intr:(9)      |1853786.765 ms |  7939307 | avg:   0.043 ms | max:  14.061 ms | sum:340332.470 ms |
  chrome:(8)            |  32196.003 ms |   213646 | avg:   0.257 ms | max:  16.000 ms | sum:54881.879 ms |
  TOTAL:                |2329645.888 ms | 23157228 |                 |       16.475 ms |   1340188.243 ms |

nomodeset
stock
  massive_intr:(9)      |1618277.030 ms |  1762657 | avg:   0.339 ms | max:  15.149 ms | sum:597323.918 ms |
  chrome:(17)           |  44370.796 ms |   130519 | avg:   0.385 ms | max:  15.948 ms | sum:50216.546 ms |
  TOTAL:                |2366021.752 ms |  6242921 |                 |       20.655 ms |   1633058.780 ms |
evdf
  massive_intr:(9)      |1794662.731 ms |   533364 | avg:   0.760 ms | max:  32.242 ms | sum:405150.225 ms |
  chrome:(16)           |  32630.283 ms |    76394 | avg:   1.276 ms | max:  25.762 ms | sum:97491.996 ms |
  TOTAL:                |2338042.935 ms |  2756565 |                 |       32.242 ms |   3230287.162 ms |
evdf+no_run_to_parity
  massive_intr:(9)      |1697254.229 ms |  1584416 | avg:   0.321 ms | max:  19.086 ms | sum:508946.014 ms |
  chrome:(16)           |  39621.161 ms |   125058 | avg:   0.500 ms | max:  18.078 ms | sum:62565.629 ms |
  TOTAL:                |2376861.778 ms |  6325320 |                 |       24.005 ms |   1714658.200 ms |

Full perf summaries below (tl-dnr unless bored), end of filename describes setup.

perf sched lat -i perf.data.full.6.4.16-eevdf.stock --sort=runtime -S 15 -T

 ----------------------------------------------------------------------------------------------------------
  Task                  |   Runtime ms  | Switches | Avg delay ms    | Max delay ms    | Sum delay ms     |
 ----------------------------------------------------------------------------------------------------------
  massive_intr:(9)      |1827981.908 ms |  9163472 | avg:   0.040 ms | max:  13.026 ms | sum:366012.231 ms |
  ThreadPoolForeg:(19)  | 171104.054 ms |  1929317 | avg:   0.101 ms | max:  19.050 ms | sum:195506.036 ms |
  Compositor:19163      |  49069.470 ms |   239950 | avg:   0.140 ms | max:  12.894 ms | sum:33621.049 ms |
  X:2146                |  46323.935 ms |   428673 | avg:   0.070 ms | max:   9.189 ms | sum:30078.222 ms |
  VizCompositorTh:19249 |  41340.942 ms |   264742 | avg:   0.155 ms | max:   8.100 ms | sum:40944.049 ms |
  chrome:(8)            |  33240.153 ms |   231301 | avg:   0.167 ms | max:   9.843 ms | sum:38606.462 ms |
  VideoFrameCompo:19257 |  21586.135 ms |   164881 | avg:   0.159 ms | max:   7.957 ms | sum:26136.556 ms |
  kwin_x11:2580         |  17878.786 ms |   197481 | avg:   0.111 ms | max:   9.070 ms | sum:21872.000 ms |
  kthreadd:(18)         |  14951.452 ms |  2671846 | avg:   0.009 ms | max:   8.010 ms | sum:23349.382 ms |
  Chrome_ChildIOT:(5)   |   7253.392 ms |   206688 | avg:   0.117 ms | max:   7.288 ms | sum:24099.171 ms |
  cyclictest:(9)        |   5707.305 ms |  1058199 | avg:   0.171 ms | max:  13.533 ms | sum:181391.599 ms |
  kworker/7:1+eve:18761 |   5201.144 ms |   957751 | avg:   0.008 ms | max:   8.909 ms | sum: 8113.339 ms |
  perf:(2)              |   5025.244 ms |     1865 | avg:   0.107 ms | max:   7.227 ms | sum:  199.488 ms |
  kworker/0:0-eve:18017 |   4967.620 ms |   916589 | avg:   0.007 ms | max:   9.728 ms | sum: 6758.811 ms |
  Media:19256           |   4964.743 ms |    83078 | avg:   0.177 ms | max:   9.663 ms | sum:14695.510 ms |
 ----------------------------------------------------------------------------------------------------------
  TOTAL:                |2307670.845 ms | 25693098 |                 |       19.050 ms |   1110835.993 ms |
 ----------------------------------------------------------------------------------------------------------

perf sched lat -i perf.data.full.6.4.16-eevdf.evdf --sort=runtime -S 15 -T

 ----------------------------------------------------------------------------------------------------------
  Task                  |   Runtime ms  | Switches | Avg delay ms    | Max delay ms    | Sum delay ms     |
 ----------------------------------------------------------------------------------------------------------
  massive_intr:(9)      |1882920.417 ms |   575899 | avg:   0.536 ms | max:  31.997 ms | sum:308728.289 ms |
  ThreadPoolForeg:(30)  | 165069.668 ms |   470614 | avg:   0.639 ms | max:  29.031 ms | sum:300696.457 ms |
  Compositor:(2)        |  46785.446 ms |    60712 | avg:   0.858 ms | max:  31.196 ms | sum:52104.985 ms |
  X:2146                |  46547.283 ms |   107721 | avg:   0.513 ms | max:  17.115 ms | sum:55250.852 ms |
  VizCompositorTh:17683 |  40856.147 ms |    60474 | avg:   0.905 ms | max:  20.163 ms | sum:54729.452 ms |
  chrome:(12)           |  33353.359 ms |    79764 | avg:   1.082 ms | max:  22.964 ms | sum:86266.324 ms |
  VideoFrameCompo:17692 |  17277.120 ms |    53230 | avg:   0.853 ms | max:  15.759 ms | sum:45393.835 ms |
  kwin_x11:2580         |  13312.455 ms |    94547 | avg:   0.302 ms | max:  15.399 ms | sum:28526.434 ms |
  Chrome_ChildIOT:(9)   |   6318.114 ms |    94165 | avg:   0.902 ms | max:  16.436 ms | sum:84971.013 ms |
  Media:17691           |   4880.461 ms |    45385 | avg:   0.843 ms | max:  13.567 ms | sum:38262.166 ms |
  cyclictest:(9)        |   3559.891 ms |   666850 | avg:   1.982 ms | max:  20.180 ms | sum:1321503.199 ms |
  kthreadd:(13)         |   3474.286 ms |   602855 | avg:   0.074 ms | max:  17.846 ms | sum:44813.758 ms |
  alsa-sink-ALC88:2088  |   2583.063 ms |    41163 | avg:   0.005 ms | max:   0.515 ms | sum:  188.379 ms |
  plasmashell:(2)       |   2362.395 ms |    19626 | avg:   1.068 ms | max:  14.794 ms | sum:20955.374 ms |
  threaded-ml:17712     |   1487.397 ms |    17958 | avg:   1.165 ms | max:  16.443 ms | sum:20926.765 ms |
 ----------------------------------------------------------------------------------------------------------
  TOTAL:                |2289006.437 ms |  5105092 |                 |       31.997 ms |   2748046.342 ms |
 ----------------------------------------------------------------------------------------------------------

perf sched lat -i perf.data.full.6.4.16-eevdf.evdf_no_run_to_parity --sort=runtime -S 15 -T

 ----------------------------------------------------------------------------------------------------------
  Task                  |   Runtime ms  | Switches | Avg delay ms    | Max delay ms    | Sum delay ms     |
 ----------------------------------------------------------------------------------------------------------
  massive_intr:(9)      |1853786.765 ms |  7939307 | avg:   0.043 ms | max:  14.061 ms | sum:340332.470 ms |
  ThreadPoolForeg:(20)  | 167239.704 ms |  2256833 | avg:   0.113 ms | max:  16.475 ms | sum:254078.291 ms |
  Compositor:18389      |  48419.605 ms |   200778 | avg:   0.230 ms | max:  12.690 ms | sum:46259.813 ms |
  X:2146                |  45209.241 ms |   333263 | avg:   0.138 ms | max:  11.095 ms | sum:45850.493 ms |
  VizCompositorTh:18464 |  40724.452 ms |   246254 | avg:   0.219 ms | max:  11.846 ms | sum:53828.807 ms |
  chrome:(8)            |  32196.003 ms |   213646 | avg:   0.257 ms | max:  16.000 ms | sum:54881.879 ms |
  VideoFrameCompo:18474 |  19577.415 ms |   180778 | avg:   0.195 ms | max:   9.806 ms | sum:35266.517 ms |
  kwin_x11:2580         |  16396.662 ms |   197846 | avg:   0.125 ms | max:   8.634 ms | sum:24647.504 ms |
  kthreadd:(15)         |   6986.487 ms |  1231177 | avg:   0.015 ms | max:   9.100 ms | sum:18079.114 ms |
  Chrome_ChildIOT:(5)   |   6502.116 ms |   190137 | avg:   0.268 ms | max:  10.421 ms | sum:50981.430 ms |
  kworker/7:0-eve:18015 |   6253.830 ms |  1161609 | avg:   0.012 ms | max:  10.946 ms | sum:13745.498 ms |
  cyclictest:(9)        |   5898.072 ms |  1071596 | avg:   0.166 ms | max:  10.758 ms | sum:177833.361 ms |
  kworker/5:2-eve:17027 |   5685.599 ms |  1055878 | avg:   0.012 ms | max:   7.236 ms | sum:12585.682 ms |
  rastertospl:18767     |   5531.070 ms |   157164 | avg:   0.024 ms | max:  11.738 ms | sum: 3703.942 ms |
  kworker/6:1-eve:17794 |   4993.173 ms |   926425 | avg:   0.013 ms | max:   7.288 ms | sum:11719.453 ms |
 ----------------------------------------------------------------------------------------------------------
  TOTAL:                |2329645.888 ms | 23157228 |                 |       16.475 ms |   1340188.243 ms |
 ----------------------------------------------------------------------------------------------------------

perf sched lat -i perf.data.full.6.4.16-eevdf.nms_stock --sort=runtime -S 15 -T

 ----------------------------------------------------------------------------------------------------------
  Task                  |   Runtime ms  | Switches | Avg delay ms    | Max delay ms    | Sum delay ms     |
 ----------------------------------------------------------------------------------------------------------
  massive_intr:(9)      |1618277.030 ms |  1762657 | avg:   0.339 ms | max:  15.149 ms | sum:597323.918 ms |
  ThreadPoolForeg:(30)  | 206450.133 ms |  1290121 | avg:   0.189 ms | max:  20.655 ms | sum:243361.803 ms |
  X:2144                | 129827.318 ms |   211518 | avg:   0.271 ms | max:  11.946 ms | sum:57368.468 ms |
  VizCompositorTh:5013  |  60322.953 ms |   125632 | avg:   0.429 ms | max:   8.758 ms | sum:53893.618 ms |
  Compositor:(2)        |  60265.270 ms |   135041 | avg:   0.362 ms | max:   9.609 ms | sum:48833.756 ms |
  chrome:(17)           |  44370.796 ms |   130519 | avg:   0.385 ms | max:  15.948 ms | sum:50216.546 ms |
  VideoFrameCompo:5021  |  24456.245 ms |    83088 | avg:   0.366 ms | max:  10.572 ms | sum:30424.287 ms |
  llvmpipe-6:(3)        |  23094.705 ms |    75617 | avg:   0.504 ms | max:   9.767 ms | sum:38106.696 ms |
  llvmpipe-5:(3)        |  22518.782 ms |    75909 | avg:   0.506 ms | max:  10.071 ms | sum:38447.875 ms |
  llvmpipe-2:(3)        |  22515.065 ms |    77387 | avg:   0.506 ms | max:  10.304 ms | sum:39181.863 ms |
  llvmpipe-4:(3)        |  22348.397 ms |    76206 | avg:   0.512 ms | max:  10.878 ms | sum:38981.961 ms |
  llvmpipe-7:(3)        |  21449.893 ms |    72673 | avg:   0.545 ms | max:  10.080 ms | sum:39585.490 ms |
  llvmpipe-1:(3)        |  21445.333 ms |    77249 | avg:   0.511 ms | max:  10.577 ms | sum:39436.002 ms |
  llvmpipe-3:(3)        |  21385.809 ms |    76409 | avg:   0.514 ms | max:   9.779 ms | sum:39295.360 ms |
  llvmpipe-0:(3)        |  18373.092 ms |    76699 | avg:   0.540 ms | max:  10.070 ms | sum:41411.515 ms |
 ----------------------------------------------------------------------------------------------------------
  TOTAL:                |2366021.752 ms |  6242921 |                 |       20.655 ms |   1633058.780 ms |
 ----------------------------------------------------------------------------------------------------------

perf sched lat -i perf.data.full.6.4.16-eevdf.nms_evdf --sort=runtime -S 15 -T

 ----------------------------------------------------------------------------------------------------------
  Task                  |   Runtime ms  | Switches | Avg delay ms    | Max delay ms    | Sum delay ms     |
 ----------------------------------------------------------------------------------------------------------
  massive_intr:(9)      |1794662.731 ms |   533364 | avg:   0.760 ms | max:  32.242 ms | sum:405150.225 ms |
  ThreadPoolForeg:(20)  | 161341.055 ms |   445425 | avg:   0.729 ms | max:  25.981 ms | sum:324736.035 ms |
  X:2144                |  91358.513 ms |    80781 | avg:   0.668 ms | max:  23.512 ms | sum:53947.202 ms |
  VizCompositorTh:5628  |  53228.085 ms |    64676 | avg:   1.041 ms | max:  22.603 ms | sum:67298.785 ms |
  Compositor:5535       |  53070.756 ms |    64927 | avg:   0.978 ms | max:  21.131 ms | sum:63507.765 ms |
  chrome:(16)           |  32630.283 ms |    76394 | avg:   1.276 ms | max:  25.762 ms | sum:97491.996 ms |
  VideoFrameCompo:5636  |  18292.351 ms |    51680 | avg:   0.939 ms | max:  26.293 ms | sum:48545.469 ms |
  llvmpipe-2:(3)        |  12897.634 ms |    36193 | avg:   1.327 ms | max:  19.412 ms | sum:48042.017 ms |
  llvmpipe-5:(3)        |  12275.682 ms |    36512 | avg:   1.329 ms | max:  13.125 ms | sum:48509.590 ms |
  llvmpipe-4:(3)        |  12196.303 ms |    36457 | avg:   1.334 ms | max:  21.460 ms | sum:48620.630 ms |
  llvmpipe-7:(3)        |  12119.861 ms |    36633 | avg:   1.324 ms | max:  17.920 ms | sum:48514.587 ms |
  llvmpipe-0:(3)        |  12106.474 ms |    37216 | avg:   1.256 ms | max:  13.919 ms | sum:46752.353 ms |
  llvmpipe-3:(3)        |  11788.836 ms |    36487 | avg:   1.344 ms | max:  13.044 ms | sum:49047.686 ms |
  llvmpipe-1:(3)        |  11630.206 ms |    36582 | avg:   1.329 ms | max:  14.691 ms | sum:48600.821 ms |
  llvmpipe-6:(3)        |  11523.446 ms |    36457 | avg:   1.353 ms | max:  16.520 ms | sum:49325.031 ms |
 ----------------------------------------------------------------------------------------------------------
  TOTAL:                |2338042.935 ms |  2756565 |                 |       32.242 ms |   3230287.162 ms |
 ----------------------------------------------------------------------------------------------------------

perf sched lat -i perf.data.full.6.4.16-eevdf.nms_evdf_no_run_to_parity --sort=runtime -S 15 -T

 ----------------------------------------------------------------------------------------------------------
  Task                  |   Runtime ms  | Switches | Avg delay ms    | Max delay ms    | Sum delay ms     |
 ----------------------------------------------------------------------------------------------------------
  massive_intr:(9)      |1697254.229 ms |  1584416 | avg:   0.321 ms | max:  19.086 ms | sum:508946.014 ms |
  ThreadPoolForeg:(21)  | 185078.035 ms |  1677435 | avg:   0.171 ms | max:  24.005 ms | sum:286473.362 ms |
  X:2144                | 119097.420 ms |   180142 | avg:   0.296 ms | max:  12.004 ms | sum:53361.641 ms |
  Compositor:6108       |  59219.138 ms |   124003 | avg:   0.470 ms | max:  10.081 ms | sum:58341.411 ms |
  VizCompositorTh:6199  |  58309.342 ms |   115937 | avg:   0.561 ms | max:  11.647 ms | sum:64992.383 ms |
  chrome:(16)           |  39621.161 ms |   125058 | avg:   0.500 ms | max:  18.078 ms | sum:62565.629 ms |
  VideoFrameCompo:6208  |  22411.289 ms |    88274 | avg:   0.439 ms | max:  15.439 ms | sum:38730.863 ms |
  llvmpipe-6:(3)        |  20137.882 ms |    70402 | avg:   0.572 ms | max:  10.372 ms | sum:40297.152 ms |
  llvmpipe-7:(3)        |  19835.717 ms |    69843 | avg:   0.575 ms | max:   9.998 ms | sum:40176.008 ms |
  llvmpipe-5:(3)        |  19401.835 ms |    69728 | avg:   0.589 ms | max:   8.997 ms | sum:41069.559 ms |
  llvmpipe-4:(3)        |  18977.364 ms |    69932 | avg:   0.591 ms | max:  13.462 ms | sum:41361.608 ms |
  llvmpipe-3:(3)        |  18459.713 ms |    69809 | avg:   0.599 ms | max:  10.459 ms | sum:41800.434 ms |
  llvmpipe-2:(3)        |  18255.747 ms |    70021 | avg:   0.604 ms | max:  11.308 ms | sum:42259.431 ms |
  llvmpipe-1:(3)        |  17637.013 ms |    69658 | avg:   0.613 ms | max:  10.963 ms | sum:42710.066 ms |
  llvmpipe-0:(3)        |  15466.278 ms |    68809 | avg:   0.648 ms | max:   9.798 ms | sum:44575.145 ms |
 ----------------------------------------------------------------------------------------------------------
  TOTAL:                |2376861.778 ms |  6325320 |                 |       24.005 ms |   1714658.200 ms |
 ----------------------------------------------------------------------------------------------------------

Peter Zijlstra Oct. 15, 2023, 10:44 a.m. UTC | #2

On Thu, Oct 12, 2023 at 10:02:13PM -0500, Youssef Esmat wrote:
> Interactive workloads see performance gains by disabling eligibility
> checks (EEVDF->EVDF). Disabling the checks reduces the number of
> context switches and delays less important work (higher deadlines/nice
> values) in favor of more important work (lower deadlines/nice values).
> 
> That said, that can add large latencies for some work loads and as the
> default is eligibility on, but allowing it to be turned off when
> beneficial.
> 
> Signed-off-by: Youssef Esmat <youssefesmat@chromium.org>
> Link: https://lore.kernel.org/lkml/CA+q576MS0-MV1Oy-eecvmYpvNT3tqxD8syzrpxQ-Zk310hvRbw@mail.gmail.com/
> ---
>  kernel/sched/fair.c     | 3 +++
>  kernel/sched/features.h | 1 +
>  2 files changed, 4 insertions(+)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index a751e552f253..16106da5a354 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -728,6 +728,9 @@ int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se)
>  	s64 avg = cfs_rq->avg_vruntime;
>  	long load = cfs_rq->avg_load;
>  
> +	if (!sched_feat(ENFORCE_ELIGIBILITY))
> +		return 1;
> +
>  	if (curr && curr->on_rq) {
>  		unsigned long weight = scale_load_down(curr->load.weight);
>  

Right.. could you pretty please try:

  git://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git sched/eevdf

as of yesterday or so.

It defaults to (EEVDF relevant features):

SCHED_FEAT(PLACE_LAG, true)
SCHED_FEAT(PLACE_DEADLINE_INITIAL, true)
SCHED_FEAT(PREEMPT_SHORT, true)
SCHED_FEAT(PLACE_SLEEPER, false)
SCHED_FEAT(GENTLE_SLEEPER, true)
SCHED_FEAT(EVDF, false)
SCHED_FEAT(DELAY_DEQUEUE, true)
SCHED_FEAT(GENTLE_DELAY, true)

If that doesn't do well enough, could you please try, in order of
preference:

2) NO_GENTLE_DELAY
3) NO_DELAY_DEQUEUE, PLACE_SLEEPER
4) NO_DELAY_DEQUEUE, PLACE_SLEEPER, NO_GENTLE_SLEEPER

I really don't like the EVDF option, and I think you'll end up
regretting using it sooner rather than later, just to make this one
benchmark you have happy.

I'm hoping the default is enough, but otherwise any of the above should
be a *much* better scheduler.

Also, bonus points if you can create us a stand alone benchmark that
captures your metric (al-la facebook's schbench) without the whole
chrome nonsense, that'd be epic.

Tor Vic Oct. 16, 2023, 1:33 p.m. UTC | #3

On 10/15/23 12:44, Peter Zijlstra wrote:
> On Thu, Oct 12, 2023 at 10:02:13PM -0500, Youssef Esmat wrote:
>> Interactive workloads see performance gains by disabling eligibility
>> checks (EEVDF->EVDF). Disabling the checks reduces the number of
>> context switches and delays less important work (higher deadlines/nice
>> values) in favor of more important work (lower deadlines/nice values).
>>
>> That said, that can add large latencies for some work loads and as the
>> default is eligibility on, but allowing it to be turned off when
>> beneficial.
>>
>> Signed-off-by: Youssef Esmat <youssefesmat@chromium.org>
>> Link: https://lore.kernel.org/lkml/CA+q576MS0-MV1Oy-eecvmYpvNT3tqxD8syzrpxQ-Zk310hvRbw@mail.gmail.com/
>> ---
>>   kernel/sched/fair.c     | 3 +++
>>   kernel/sched/features.h | 1 +
>>   2 files changed, 4 insertions(+)
>>
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index a751e552f253..16106da5a354 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -728,6 +728,9 @@ int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se)
>>   	s64 avg = cfs_rq->avg_vruntime;
>>   	long load = cfs_rq->avg_load;
>>   
>> +	if (!sched_feat(ENFORCE_ELIGIBILITY))
>> +		return 1;
>> +
>>   	if (curr && curr->on_rq) {
>>   		unsigned long weight = scale_load_down(curr->load.weight);
>>   
> 
> Right.. could you pretty please try:
> 
>    git://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git sched/eevdf
> 
> as of yesterday or so.
> 
> It defaults to (EEVDF relevant features):
> 
> SCHED_FEAT(PLACE_LAG, true)
> SCHED_FEAT(PLACE_DEADLINE_INITIAL, true)
> SCHED_FEAT(PREEMPT_SHORT, true)
> SCHED_FEAT(PLACE_SLEEPER, false)
> SCHED_FEAT(GENTLE_SLEEPER, true)
> SCHED_FEAT(EVDF, false)
> SCHED_FEAT(DELAY_DEQUEUE, true)
> SCHED_FEAT(GENTLE_DELAY, true)
> 
> If that doesn't do well enough, could you please try, in order of
> preference:
> 
> 2) NO_GENTLE_DELAY
> 3) NO_DELAY_DEQUEUE, PLACE_SLEEPER
> 4) NO_DELAY_DEQUEUE, PLACE_SLEEPER, NO_GENTLE_SLEEPER

I'm very interested in this scheduler stuff, but I know nothing about 
the code.

Still, I ran some very quick benchmarks on a dual-core Skylake laptop 
running 6.6-rc6.
Base slice is 5 ms.

1) Without the recent patches from Peter's tree
2) With patches, default features
3) With patches, NO_GENTLE_DELAY
4) With patches, NO_DELAY_DEQUEUE + PLACE_SLEEPER
5) With patches, like 4) + NO_GENTLE_SLEEPER
6) With patches, like 5) + EVDF

   $ perf stat -r 7 -e cs,migrations,cache-misses,branch-misses -- perf 
bench sched messaging -g 20 -l 1000 -p

test  | seconds | cs   | migrations | cache miss | branch miss |
------|---------|------|------------|------------|-------------|
1)    |  2,90   | 192K |   6,7K     |   99M      |   60M       |
2)    |  2,97   | 226K |   6,9K     |  102M      |   61M       |
3)    |  3,00   | 247K |   6,9K     |  108M      |   62M       |
4)    |  2,92   | 182K |   7,2K     |  101M      |   60M       |
5)    |  2,94   | 203K |   6,8K     |  101M      |   60M       |
6)    |  2,79   |  84K |   6,4K     |   94M      |   57M       |


   $ stress-ng --bsearch 2 --matrix 2 --matrix-method prod --timeout 30 
--metrics-brief [results in bogo ops/s]

test  | bsearch | matrix |
------|---------|--------|
1)    |  392    |  588   |
2)    |  512    |  688   |
3)    |  512    |  663   |
4)    |  512    |  688   |
5)    |  511    |  686   |
6)    |  510    |  655   |

--

I don't know if this info is useful enough for you scheduler people, but 
I hope it helps.

Cheers,
Tor

> 
> I really don't like the EVDF option, and I think you'll end up
> regretting using it sooner rather than later, just to make this one
> benchmark you have happy.
> 
> I'm hoping the default is enough, but otherwise any of the above should
> be a *much* better scheduler.
> 
> Also, bonus points if you can create us a stand alone benchmark that
> captures your metric (al-la facebook's schbench) without the whole
> chrome nonsense, that'd be epic.
>

Steven Rostedt Oct. 16, 2023, 3:28 p.m. UTC | #4

On Sun, 15 Oct 2023 12:44:28 +0200
Peter Zijlstra <peterz@infradead.org> wrote:

> Right.. could you pretty please try:
> 
>   git://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git sched/eevdf
> 
> as of yesterday or so.
> 
> It defaults to (EEVDF relevant features):
> 
> SCHED_FEAT(PLACE_LAG, true)
> SCHED_FEAT(PLACE_DEADLINE_INITIAL, true)
> SCHED_FEAT(PREEMPT_SHORT, true)
> SCHED_FEAT(PLACE_SLEEPER, false)
> SCHED_FEAT(GENTLE_SLEEPER, true)
> SCHED_FEAT(EVDF, false)
> SCHED_FEAT(DELAY_DEQUEUE, true)
> SCHED_FEAT(GENTLE_DELAY, true)
> 
> If that doesn't do well enough, could you please try, in order of
> preference:
> 
> 2) NO_GENTLE_DELAY
> 3) NO_DELAY_DEQUEUE, PLACE_SLEEPER
> 4) NO_DELAY_DEQUEUE, PLACE_SLEEPER, NO_GENTLE_SLEEPER

Thanks Peter, we'll give this a try.

> 
> I really don't like the EVDF option, and I think you'll end up
> regretting using it sooner rather than later, just to make this one
> benchmark you have happy.

Note, the benchmark we use is very close to real world settings that we
care about. And if we were to go further with Youssef's feature, we would
test it by sending it out to 1% of our user base, then 2%, 5%, and so on,
with a lot more feedback analysis going on. If it were to cause any
regressions, it would likely be noticed during this process, and be able to
back out any changes.

The main point is, our testing is not around any single benchmark that we
are trying to make happy. We really are looking at what makes the user base
run better in the real world.

> 
> I'm hoping the default is enough, but otherwise any of the above should
> be a *much* better scheduler.
> 
> Also, bonus points if you can create us a stand alone benchmark that
> captures your metric (al-la facebook's schbench) without the whole
> chrome nonsense, that'd be epic.

As I stated above. We don't really care about any one benchmark, but our
focus is on our user base. It's not as simple as what facebook would have,
as they are server focused and have a lot more information to test with. We
are more focused on the quality of chromebooks for kids in school, which is
much more difficult to analyze ;-)

What we could do, is give you a way to have access to run our benchmarks in
our infrastructure if you want to test anything in particular. Would you be
interested in that?

-- Steve

Peter Zijlstra Oct. 16, 2023, 4:52 p.m. UTC | #5

On Mon, Oct 16, 2023 at 11:28:51AM -0400, Steven Rostedt wrote:

> The main point is, our testing is not around any single benchmark that we
> are trying to make happy. We really are looking at what makes the user base
> run better in the real world.

This is a key (har-har) performance indicator for you guys though, I've
seen it mentioned before (with the core-scheduling crud IIRC).

As such it would be good to capture in a stand alone program and share.

diff mbox series

Patch

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index a751e552f253..16106da5a354 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -728,6 +728,9 @@  int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se)
 	s64 avg = cfs_rq->avg_vruntime;
 	long load = cfs_rq->avg_load;
 
+	if (!sched_feat(ENFORCE_ELIGIBILITY))
+		return 1;
+
 	if (curr && curr->on_rq) {
 		unsigned long weight = scale_load_down(curr->load.weight);
 
diff --git a/kernel/sched/features.h b/kernel/sched/features.h
index f770168230ae..84e38a0045b7 100644
--- a/kernel/sched/features.h
+++ b/kernel/sched/features.h
@@ -7,6 +7,7 @@ 
 SCHED_FEAT(PLACE_LAG, true)
 SCHED_FEAT(PLACE_DEADLINE_INITIAL, true)
 SCHED_FEAT(RUN_TO_PARITY, true)
+SCHED_FEAT(ENFORCE_ELIGIBILITY, true)
 
 /*
  * Prefer to schedule the task we woke last (assuming it failed