Re: [PATCH] sched: next buddy hint on sleep and preempt path

[Paul Turner <pjt@google.com>]

On Tue, Mar 1, 2011 at 3:33 PM, Venkatesh Pallipadi <venki@google.com> wrote:
> When a task in a taskgroup sleeps, pick_next_task starts all the way back at
> the root and picks the task/taskgroup with the min vruntime across all
> runnable tasks. But, when there are many frequently sleeping tasks
> across different taskgroups, it makes better sense to stay with same taskgroup
> for its slice period (or until all tasks in the taskgroup sleeps) instead of
> switching cross taskgroup on each sleep after a short runtime.
> This helps specifically where taskgroups corresponds to a process with
> multiple threads. The change reduces the number of CR3 switches in this case.
>
> Example:
> Two taskgroups with 2 threads each which are running for 2ms and
> sleeping for 1ms. Looking at sched:sched_switch shows -
>
> BEFORE: taskgroup_1 threads [5004, 5005], taskgroup_2 threads [5016, 5017]
>      cpu-soaker-5004  [003]  3683.391089
>      cpu-soaker-5016  [003]  3683.393106
>      cpu-soaker-5005  [003]  3683.395119
>      cpu-soaker-5017  [003]  3683.397130
>      cpu-soaker-5004  [003]  3683.399143
>      cpu-soaker-5016  [003]  3683.401155
>      cpu-soaker-5005  [003]  3683.403168
>      cpu-soaker-5017  [003]  3683.405170
>
> AFTER: taskgroup_1 threads [21890, 21891], taskgroup_2 threads [21934, 21935]
>      cpu-soaker-21890 [003]   865.895494
>      cpu-soaker-21935 [003]   865.897506
>      cpu-soaker-21934 [003]   865.899520
>      cpu-soaker-21935 [003]   865.901532
>      cpu-soaker-21934 [003]   865.903543
>      cpu-soaker-21935 [003]   865.905546
>      cpu-soaker-21891 [003]   865.907548
>      cpu-soaker-21890 [003]   865.909560
>      cpu-soaker-21891 [003]   865.911571
>      cpu-soaker-21890 [003]   865.913582
>      cpu-soaker-21891 [003]   865.915594
>      cpu-soaker-21934 [003]   865.917606
>
> Similar problem is there when there are multiple taskgroups and say a task A
> preempts currently running task B of taskgroup_1. On schedule, pick_next_task
> can pick an unrelated task on taskgroup_2. Here it would be better to give some
> preference to task B on pick_next_task.
>
> A simple (may be extreme case) benchmark I tried was tbench with 2 tbench
> client processes with 2 threads each running on a single CPU. Avg throughput
> across 5 50 sec runs was -
> BEFORE: 105.84 MB/sec
> AFTER: 112.42 MB/sec
>
> Signed-off-by: Venkatesh Pallipadi <venki@google.com>
> ---
>  kernel/sched_fair.c |   20 ++++++++++++++++++--
>  1 files changed, 18 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
> index 3a88dee..36e8f02 100644
> --- a/kernel/sched_fair.c
> +++ b/kernel/sched_fair.c
> @@ -1339,6 +1339,8 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
>        hrtick_update(rq);
>  }
>
> +static void set_next_buddy(struct sched_entity *se);
> +
>  /*
>  * The dequeue_task method is called before nr_running is
>  * decreased. We remove the task from the rbtree and
> @@ -1348,14 +1350,22 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
>  {
>        struct cfs_rq *cfs_rq;
>        struct sched_entity *se = &p->se;
> +       int task_flags = flags;

simpler: int voluntary = flags & DEQUEUE_SLEEP;
>
>        for_each_sched_entity(se) {
>                cfs_rq = cfs_rq_of(se);
>                dequeue_entity(cfs_rq, se, flags);
>
>                /* Don't dequeue parent if it has other entities besides us */
> -               if (cfs_rq->load.weight)
> +               if (cfs_rq->load.weight) {
> +                       /*
> +                        * Bias pick_next to pick a task from this cfs_rq, as
> +                        * p is sleeping when it is within its sched_slice.
> +                        */
> +                       if (task_flags & DEQUEUE_SLEEP && se->parent)
> +                               set_next_buddy(se->parent);

re-using the last_buddy would seem like a more natural fit here; also
doesn't have a clobber race with a wakeup

>                        break;
> +               }
>                flags |= DEQUEUE_SLEEP;
>        }
>
> @@ -1887,8 +1897,14 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_
>        update_curr(cfs_rq);
>        find_matching_se(&se, &pse);
>        BUG_ON(!pse);
> -       if (wakeup_preempt_entity(se, pse) == 1)
> +       if (wakeup_preempt_entity(se, pse) == 1) {
> +               /*
> +                * Bias pick_next to pick the sched entity that is
> +                * triggering this preemption.
> +                */
> +               set_next_buddy(pse);

this probably wants some sort of unification with the scale-based next
buddy above

>                goto preempt;
> +       }
>
>        return;
>
> --
> 1.7.3.1
>
>