On 03/25/2013 01:24 PM, Michael Wang wrote:
> Recently testing show that wake-affine stuff cause regression on pgbench, the
> hiding rat was finally catched out.
>
> wake-affine stuff is always trying to pull wakee close to waker, by theory,
> this will benefit us if waker's cpu cached hot data for wakee, or the extreme
> ping-pong case.
>
> However, the whole stuff is somewhat blindly, there is no examining on the
> relationship between waker and wakee, and since the stuff itself
> is time-consuming, some workload suffered, pgbench is just the one who
> has been found.
>
> Thus, throttle the wake-affine stuff for such workload is necessary.
>
> This patch introduced a new knob 'sysctl_sched_wake_affine_interval' with the
> default value 1ms, which means wake-affine stuff only effect once per 1ms,
> which
> usually the minimum balance interval (the idea is that the rapid of
> wake-affine
> should lower than the rapid of load-balance at least).
>
> By turning the new knob, those workload who suffered will have the chance to
> stop the regression.
My recently testing show this idea still works well after the
wake-affine itself was optimized.
Throttle seems to be an easy and efficient way to stop the regression of
pgbench caused by wake-affine stuff, should we adopt this approach?
Please let me know if there are still any concerns ;-)
Regards,
Michael Wang
>
> Test:
> Test with 12 cpu X86 server and tip 3.9.0-rc2.
>
> Default 1ms interval bring limited performance improvement(<5%) for
> pgbench, significant improvement start to show when turning the
> knob to 100ms.
>
> original 100ms
>
> | db_size | clients | tps | | tps |
> +---------+---------+-------+ +-------+
> | 21 MB | 1 | 10572 | | 10675 |
> | 21 MB | 2 | 21275 | | 21228 |
> | 21 MB | 4 | 41866 | | 41946 |
> | 21 MB | 8 | 53931 | | 55176 |
> | 21 MB | 12 | 50956 | | 54457 | +6.87%
> | 21 MB | 16 | 49911 | | 55468 | +11.11%
> | 21 MB | 24 | 46046 | | 56446 | +22.59%
> | 21 MB | 32 | 43405 | | 55177 | +27.12%
> | 7483 MB | 1 | 7734 | | 7721 |
> | 7483 MB | 2 | 19375 | | 19277 |
> | 7483 MB | 4 | 37408 | | 37685 |
> | 7483 MB | 8 | 49033 | | 49152 |
> | 7483 MB | 12 | 45525 | | 49241 | +8.16%
> | 7483 MB | 16 | 45731 | | 51425 | +12.45%
> | 7483 MB | 24 | 41533 | | 52349 | +26.04%
> | 7483 MB | 32 | 36370 | | 51022 | +40.28%
> | 15 GB | 1 | 7576 | | 7422 |
> | 15 GB | 2 | 19157 | | 19176 |
> | 15 GB | 4 | 37285 | | 36982 |
> | 15 GB | 8 | 48718 | | 48413 |
> | 15 GB | 12 | 45167 | | 48497 | +7.37%
> | 15 GB | 16 | 45270 | | 51276 | +13.27%
> | 15 GB | 24 | 40984 | | 51628 | +25.97%
> | 15 GB | 32 | 35918 | | 51060 | +42.16%
>
> Suggested-by: Peter Zijlstra <pet...@infradead.org>
> Signed-off-by: Michael Wang <wang...@linux.vnet.ibm.com>
> ---
> include/linux/sched.h | 5 +++++
> kernel/sched/fair.c | 33 ++++++++++++++++++++++++++++++++-
> kernel/sysctl.c | 10 ++++++++++
> 3 files changed, 47 insertions(+), 1 deletions(-)
>
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index d35d2b6..e9efd3a 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -1197,6 +1197,10 @@ enum perf_event_task_context {
> perf_nr_task_contexts,
> };
>
> +#ifdef CONFIG_SMP
> +extern unsigned int sysctl_sched_wake_affine_interval;
> +#endif
> +
> struct task_struct {
> volatile long state; /* -1 unrunnable, 0 runnable, >0 stopped */
> void *stack;
> @@ -1207,6 +1211,7 @@ struct task_struct {
> #ifdef CONFIG_SMP
> struct llist_node wake_entry;
> int on_cpu;
> + unsigned long next_wake_affine;
> #endif
> int on_rq;
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 7a33e59..00d7f45 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -3087,6 +3087,22 @@ static inline unsigned long effective_load(struct
> task_group *tg, int cpu,
>
> #endif
>
> +/*
> + * Default is 1ms, to prevent the wake_affine() stuff working too frequently.
> + */
> +unsigned int sysctl_sched_wake_affine_interval = 1U;
> +
> +static inline int wake_affine_throttled(struct task_struct *p)
> +{
> + return time_before(jiffies, p->next_wake_affine);
> +}
> +
> +static inline void wake_affine_throttle(struct task_struct *p)
> +{
> + p->next_wake_affine = jiffies +
> + msecs_to_jiffies(sysctl_sched_wake_affine_interval);
> +}
> +
> static int wake_affine(struct sched_domain *sd, struct task_struct *p, int
> sync)
> {
> s64 this_load, load;
> @@ -3096,6 +3112,9 @@ static int wake_affine(struct sched_domain *sd, struct
> task_struct *p, int sync)
> unsigned long weight;
> int balanced;
>
> + if (wake_affine_throttled(p))
> + return 0;
> +
> idx = sd->wake_idx;
> this_cpu = smp_processor_id();
> prev_cpu = task_cpu(p);
> @@ -3342,8 +3361,20 @@ select_task_rq_fair(struct task_struct *p, int
> sd_flag, int wake_flags)
> }
>
> if (affine_sd) {
> - if (cpu != prev_cpu && wake_affine(affine_sd, p, sync))
> + if (cpu != prev_cpu && wake_affine(affine_sd, p, sync)) {
> + /*
> + * wake_affine() stuff try to pull wakee to the cpu
> + * around waker, this will benefit us if the data
> + * cached on waker cpu is hot for wakee, or the extreme
> + * ping-pong case.
> + *
> + * However, do such blindly work too frequently will
> + * cause regression to some workload, thus, each time
> + * when wake_affine() succeed, throttle it for a while.
> + */
> + wake_affine_throttle(p);
> prev_cpu = cpu;
> + }
>
> new_cpu = select_idle_sibling(p, prev_cpu);
> goto unlock;
> diff --git a/kernel/sysctl.c b/kernel/sysctl.c
> index afc1dc6..6b798b6 100644
> --- a/kernel/sysctl.c
> +++ b/kernel/sysctl.c
> @@ -437,6 +437,16 @@ static struct ctl_table kern_table[] = {
> .extra1 = &one,
> },
> #endif
> +#ifdef CONFIG_SMP
> + {
> + .procname = "sched_wake_affine_interval",
> + .data = &sysctl_sched_wake_affine_interval,
> + .maxlen = sizeof(unsigned int),
> + .mode = 0644,
> + .proc_handler = proc_dointvec_minmax,
> + .extra1 = &one,
> + },
> +#endif
> #ifdef CONFIG_PROVE_LOCKING
> {
> .procname = "prove_locking",
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
