Re: [PATCH] xen/sched_credit: Use delay to control scheduling frequency

[Keir Fraser <keir@xen.org>]

On 24/01/2012 12:01, "George Dunlap" <George.Dunlap@eu.citrix.com> wrote:

> Ping?

It's been in the tree for a week.

> On Tue, Jan 10, 2012 at 10:05 AM, George Dunlap
> <George.Dunlap@eu.citrix.com> wrote:
>> On Mon, Jan 9, 2012 at 10:22 AM, Hui Lv <hui.lv@intel.com> wrote:
>>> Updated the warning sentence for ratelimit_us.
>>> 
>>> This patch can improve Xen performance:
>>> 1. Basically, the "delay method" can achieve 11% overall performance boost
>>> for SPECvirt than original credit scheduler.
>>> 2. We have tried 1ms delay and 10ms delay, there is no big difference
>>> between these two configurations. (1ms is enough to achieve a good
>>> performance)
>>> 3. We have compared different load level response time/latency (low, high,
>>> peak), "delay method" didn't bring very much response time increase.
>>> 4. 1ms delay can reduce 30% context switch at peak performance, where
>>> produces the benefits. (int sched_ratelimit_us = 1000 is the recommended
>>> setting)
>>> 
>>> Signed-off-by: Hui Lv <hui.lv@intel.com>
>>> Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>
>>> 
>>> Acked-by: George Dunlap <george.dunlap@eu.citrix.com>
>> 
>> Just confirming this:
>> Acked-by: George Dunlap <george.dunlap@eu.citrix.com>
>> 
>> Thanks, Hui.
>>  -George
>> 
>>> 
>>> diff -r a4bff36780a3 -r fe8d0ca867aa xen/common/sched_credit.c
>>> --- a/xen/common/sched_credit.c Fri Dec 16 18:46:27 2011 +0000
>>> +++ b/xen/common/sched_credit.c Mon Jan 09 05:21:35 2012 -0500
>>> @@ -172,6 +172,7 @@ struct csched_private {
>>>     uint32_t credit;
>>>     int credit_balance;
>>>     uint32_t runq_sort;
>>> +    unsigned ratelimit_us;
>>>     /* Period of master and tick in milliseconds */
>>>     unsigned tslice_ms, tick_period_us, ticks_per_tslice;
>>>     unsigned credits_per_tslice;
>>> @@ -1297,10 +1298,15 @@ csched_schedule(
>>>     struct csched_private *prv = CSCHED_PRIV(ops);
>>>     struct csched_vcpu *snext;
>>>     struct task_slice ret;
>>> +    s_time_t runtime, tslice;
>>> 
>>>     CSCHED_STAT_CRANK(schedule);
>>>     CSCHED_VCPU_CHECK(current);
>>> 
>>> +    runtime = now - current->runstate.state_entry_time;
>>> +    if ( runtime < 0 ) /* Does this ever happen? */
>>> +        runtime = 0;
>>> +
>>>     if ( !is_idle_vcpu(scurr->vcpu) )
>>>     {
>>>         /* Update credits of a non-idle VCPU. */
>>> @@ -1313,6 +1319,35 @@ csched_schedule(
>>>         scurr->pri = CSCHED_PRI_IDLE;
>>>     }
>>> 
>>> +    /* Choices, choices:
>>> +     * - If we have a tasklet, we need to run the idle vcpu no matter what.
>>> +     * - If sched rate limiting is in effect, and the current vcpu has
>>> +     *   run for less than that amount of time, continue the current one,
>>> +     *   but with a shorter timeslice and return it immediately
>>> +     * - Otherwise, chose the one with the highest priority (which may
>>> +     *   be the one currently running)
>>> +     * - If the currently running one is TS_OVER, see if there
>>> +     *   is a higher priority one waiting on the runqueue of another
>>> +     *   cpu and steal it.
>>> +     */
>>> +
>>> +    /* If we have schedule rate limiting enabled, check to see
>>> +     * how long we've run for. */
>>> +    if ( !tasklet_work_scheduled
>>> +         && prv->ratelimit_us
>>> +         && vcpu_runnable(current)
>>> +         && !is_idle_vcpu(current)
>>> +         && runtime < MICROSECS(prv->ratelimit_us) )
>>> +    {
>>> +        snext = scurr;
>>> +        snext->start_time += now;
>>> +        perfc_incr(delay_ms);
>>> +        tslice = MICROSECS(prv->ratelimit_us);
>>> +        ret.migrated = 0;
>>> +        goto out;
>>> +    }
>>> +    tslice = MILLISECS(prv->tslice_ms);
>>> +
>>>     /*
>>>      * Select next runnable local VCPU (ie top of local runq)
>>>      */
>>> @@ -1367,11 +1402,12 @@ csched_schedule(
>>>     if ( !is_idle_vcpu(snext->vcpu) )
>>>         snext->start_time += now;
>>> 
>>> +out:
>>>     /*
>>>      * Return task to run next...
>>>      */
>>>     ret.time = (is_idle_vcpu(snext->vcpu) ?
>>> -                -1 : MILLISECS(prv->tslice_ms));
>>> +                -1 : tslice);
>>>     ret.task = snext->vcpu;
>>> 
>>>     CSCHED_VCPU_CHECK(ret.task);
>>> @@ -1533,6 +1569,15 @@ csched_init(struct scheduler *ops)
>>>     prv->tick_period_us = prv->tslice_ms * 1000 / prv->ticks_per_tslice;
>>>     prv->credits_per_tslice = CSCHED_CREDITS_PER_MSEC * prv->tslice_ms;
>>> 
>>> +    if ( MICROSECS(sched_ratelimit_us) > MILLISECS(sched_credit_tslice_ms)
>>> )
>>> +    {
>>> +        printk("WARNING: sched_ratelimit_us >"
>>> +               "sched_credit_tslice_ms is undefined\n"
>>> +               "Setting ratelimit_us to 1000 * tslice_ms\n");
>>> +        prv->ratelimit_us = 1000 * prv->tslice_ms;
>>> +    }
>>> +    else
>>> +        prv->ratelimit_us = sched_ratelimit_us;
>>>     return 0;
>>>  }
>>> 
>>> diff -r a4bff36780a3 -r fe8d0ca867aa xen/common/schedule.c
>>> --- a/xen/common/schedule.c     Fri Dec 16 18:46:27 2011 +0000
>>> +++ b/xen/common/schedule.c     Mon Jan 09 05:21:35 2012 -0500
>>> @@ -47,6 +47,11 @@ string_param("sched", opt_sched);
>>>  bool_t sched_smt_power_savings = 0;
>>>  boolean_param("sched_smt_power_savings", sched_smt_power_savings);
>>> 
>>> +/* Default scheduling rate limit: 1ms
>>> + * The behavior when sched_ratelimit_us is greater than
>>> sched_credit_tslice_ms is undefined
>>> + * */
>>> +int sched_ratelimit_us = 1000;
>>> +integer_param("sched_ratelimit_us", sched_ratelimit_us);
>>>  /* Various timer handlers. */
>>>  static void s_timer_fn(void *unused);
>>>  static void vcpu_periodic_timer_fn(void *data);
>>> diff -r a4bff36780a3 -r fe8d0ca867aa xen/include/xen/perfc_defn.h
>>> --- a/xen/include/xen/perfc_defn.h      Fri Dec 16 18:46:27 2011 +0000
>>> +++ b/xen/include/xen/perfc_defn.h      Mon Jan 09 05:21:35 2012 -0500
>>> @@ -16,6 +16,7 @@ PERFCOUNTER(sched_irq,              "sch
>>>  PERFCOUNTER(sched_run,              "sched: runs through scheduler")
>>>  PERFCOUNTER(sched_ctx,              "sched: context switches")
>>> 
>>> +PERFCOUNTER(delay_ms,               "csched: delay")
>>>  PERFCOUNTER(vcpu_check,             "csched: vcpu_check")
>>>  PERFCOUNTER(schedule,               "csched: schedule")
>>>  PERFCOUNTER(acct_run,               "csched: acct_run")
>>> diff -r a4bff36780a3 -r fe8d0ca867aa xen/include/xen/sched-if.h
>>> --- a/xen/include/xen/sched-if.h        Fri Dec 16 18:46:27 2011 +0000
>>> +++ b/xen/include/xen/sched-if.h        Mon Jan 09 05:21:35 2012 -0500
>>> @@ -16,6 +16,11 @@ extern struct cpupool *cpupool0;
>>>  /* cpus currently in no cpupool */
>>>  extern cpumask_t cpupool_free_cpus;
>>> 
>>> +/* Scheduler generic parameters
>>> + * */
>>> +extern int sched_ratelimit_us;
>>> +
>>> +
>>>  /*
>>>  * In order to allow a scheduler to remap the lock->cpu mapping,
>>>  * we have a per-cpu pointer, along with a pre-allocated set of
>>> 
>>> _______________________________________________
>>> Xen-devel mailing list
>>> Xen-devel@lists.xensource.com
>>> http://lists.xensource.com/xen-devel