Re: [PATCH v3 0/5] Scan for an idle sibling in a single pass

[Vincent Guittot]

On Tue, 19 Jan 2021 at 12:22, Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> wrote:
>
> Changelog since v2
> o Remove unnecessary parameters
> o Update nr during scan only when scanning for cpus

Hi Mel,

I haven't looked at your previous version mainly because I'm chasing a
performance regression on v5.11-rcx which prevents me from testing the
impact of your patchset on my !SMT2 system.
Will do this as soon as this problem is fixed

>
> Changlog since v1
> o Move extern declaration to header for coding style
> o Remove unnecessary parameter from __select_idle_cpu
>
> This series of 5 patches reposts three patches from Peter entitled
> "select_idle_sibling() wreckage". It only scans the runqueues in a single
> pass when searching for an idle sibling.
>
> Two patches from Peter were dropped. The first patch altered how scan
> depth was calculated. Scan depth deletion is a random number generator
> with two major limitations. The avg_idle time is based on the time
> between a CPU going idle and being woken up clamped approximately by
> 2*sysctl_sched_migration_cost.  This is difficult to compare in a sensible
> fashion to avg_scan_cost. The second issue is that only the avg_scan_cost
> of scan failures is recorded and it does not decay.  This requires deeper
> surgery that would justify a patch on its own although Peter notes that
> https://lkml.kernel.org/r/20180530143105.977759909@xxxxxxxxxxxxx is
> potentially useful for an alternative avg_idle metric.
>
> The second patch dropped converted the idle core scan throttling
> mechanism to SIS_PROP. While this would unify the throttling of core
> and CPU scanning, it was not free of regressions and has_idle_cores is
> a fairly effective throttling mechanism with the caveat that it can have
> a lot of false positives for workloads like hackbench.
>
> Peter's series tried to solve three problems at once, this subset addresses
> one problem. As with anything select_idle_sibling, it's a mix of wins and
> losses but won more than it lost across a range of workloads and machines.
>
>  kernel/sched/core.c     |  18 +++--
>  kernel/sched/fair.c     | 161 ++++++++++++++++++++--------------------
>  kernel/sched/features.h |   1 -
>  kernel/sched/sched.h    |   2 +
>  4 files changed, 95 insertions(+), 87 deletions(-)
>
> --
> 2.26.2
>