Re: [LKP] [sched/fair] 070f5e860e: reaim.jobs_per_min -10.5% regression

[Xing Zhengjun]

On 6/18/2020 8:35 PM, Vincent Guittot wrote:
> On Thu, 18 Jun 2020 at 04:45, Xing Zhengjun
> <zhengjun.xing@xxxxxxxxxxxxxxx> wrote:
> >
>
> > > > This bench forks a new thread for each and every new step. But a newly forked
> > > > threads start with a load_avg and a runnable_avg set to max whereas the threads
> > > > are running shortly before exiting. This makes the CPU to be set overloaded in
> > > > some case whereas it isn't.
> > > >
> > > > Could you try the patch below ?
> > > > It fixes the problem on my setup (I have finally been able to reproduce the problem)
> > > >
> > > > ---
> > > >    kernel/sched/fair.c | 2 +-
> > > >    1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > > > index 0aeffff62807..b33a4a9e1491 100644
> > > > --- a/kernel/sched/fair.c
> > > > +++ b/kernel/sched/fair.c
> > > > @@ -807,7 +807,7 @@ void post_init_entity_util_avg(struct task_struct *p)
> > > >               }
> > > >       }
> > > >
> > > > -    sa->runnable_avg = cpu_scale;
> > > > +    sa->runnable_avg = sa->util_avg;
> > > >
> > > >       if (p->sched_class != &fair_sched_class) {
> > > >               /*
> > > > --
> > > > 2.17.1
> > >
> > > The patch above tries to move back to the group in the same classification as
> > > before but this could harm other benchmarks.
> > >
> > > There is another way to fix this by easing the migration of task in the case
> > > of migrate_util imbalance.
> > >
> > > Could you also try the patch below instead of the one above ?
> > >
> > > ---
> > >    kernel/sched/fair.c | 3 ++-
> > >    1 file changed, 2 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > > index 0aeffff62807..fcaf66c4d086 100644
> > > --- a/kernel/sched/fair.c
> > > +++ b/kernel/sched/fair.c
> > > @@ -7753,7 +7753,8 @@ static int detach_tasks(struct lb_env *env)
> > >                case migrate_util:
> > >                        util = task_util_est(p);
> > >
> > > -                     if (util > env->imbalance)
> > > +                     if (util/2 > env->imbalance &&
> > > +                         env->sd->nr_balance_failed <= env->sd->cache_nice_tries)
> > >                                goto next;
> > >
> > >                        env->imbalance -= util;
> > > --
> > > 2.17.1
> >
> > I apply the patch based on v5.7, the test result is as the following:
> >
> > =========================================================================================
> > tbox_group/testcase/rootfs/kconfig/compiler/runtime/nr_task/debug-setup/test/cpufreq_governor/ucode:
> >
> > lkp-ivb-d04/reaim/debian-x86_64-20191114.cgz/x86_64-rhel-7.6/gcc-7/300s/100%/test/five_sec/performance/0x21
> >
> > commit:
> >     9f68395333ad7f5bfe2f83473fed363d4229f11c
> >     070f5e860ee2bf588c99ef7b4c202451faa48236
> >     v5.7
> >     69c81543653bf5f2c7105086502889fa019c15cb  (the test patch)
> >
> > 9f68395333ad7f5b 070f5e860ee2bf588c99ef7b4c2                        v5.7
> > 69c81543653bf5f2c7105086502
> > ---------------- --------------------------- ---------------------------
> > ---------------------------
> >            %stddev     %change         %stddev     %change
> > %stddev     %change         %stddev
> >                \          |                \          |                \
> >           |                \
> >         0.69           -10.3%       0.62            -9.1%       0.62
> >         -7.6%       0.63        reaim.child_systime
> >         0.62            -1.0%       0.61            +0.5%       0.62
> >         +1.9%       0.63        reaim.child_utime
> >        66870           -10.0%      60187            -7.6%      61787
> >         -5.9%      62947        reaim.jobs_per_min
>
> There is an improvement but not at the same level as on my setup.
> I'm not sure with patch you tested here. Is it the last one that
> modify detach_tasks() or the previous one that modify
> post_init_entity_util_avg() ?
It is the last one that modify detach_tasks().

> Could you also try the other one ? Both patches were improving results
> on y setup but the behavior doesn't seem to be the same on your setup.

The test result for the other one has been sent in another mail.

> >        16717           -10.0%      15046            -7.6%      15446
> >         -5.9%      15736        reaim.jobs_per_min_child
> >        97.84            -1.1%      96.75            -0.4%      97.43
> >         -0.4%      97.47        reaim.jti
> >        72000           -10.8%      64216            -8.3%      66000
> >         -5.7%      67885        reaim.max_jobs_per_min
> >         0.36           +10.6%       0.40            +7.8%       0.39
> >         +6.0%       0.38        reaim.parent_time
> >         1.58 Â  2%     +71.0%       2.70 Â  2%     +26.9%       2.01 Â
> > 2%     +23.6%       1.95 Â  3%  reaim.std_dev_percent
> >         0.00 Â  5%    +110.4%       0.01 Â  3%     +48.8%       0.01 Â
> > 7%     +43.2%       0.01 Â  5%  reaim.std_dev_time
> >        50800            -2.4%      49600            -1.6%      50000
> >         -0.8%      50400        reaim.workload
> >
> >
> > > >
> ...
> > > > > > > --
> > > > > > > Zhengjun Xing
> > > > >
> > > > > --
> > > > > Zhengjun Xing
> >
> > --
> > Zhengjun Xing

--
Zhengjun Xing