[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH] xen,credit1: Add variable timeslice



Ping?

On Thu, Sep 1, 2011 at 4:30 PM, George Dunlap
<george.dunlap@xxxxxxxxxxxxx> wrote:
> Add a xen command-line parameter, sched_credit_tslice_ms,
> to set the timeslice of the credit1 scheduler.
>
> Signed-off-by: George Dunlap <george.dunlap@xxxxxxxxxxxxx>
>
> diff -r 4a4882df5649 -r 782284c5b1bc xen/common/sched_credit.c
> --- a/xen/common/sched_credit.c Wed Aug 31 15:23:49 2011 +0100
> +++ b/xen/common/sched_credit.c Thu Sep 01 16:29:50 2011 +0100
> @@ -41,15 +41,9 @@
>  */
>  #define CSCHED_DEFAULT_WEIGHT       256
>  #define CSCHED_TICKS_PER_TSLICE     3
> -#define CSCHED_TICKS_PER_ACCT       3
> -#define CSCHED_MSECS_PER_TICK       10
> -#define CSCHED_MSECS_PER_TSLICE     \
> -    (CSCHED_MSECS_PER_TICK * CSCHED_TICKS_PER_TSLICE)
> +/* Default timeslice: 30ms */
> +#define CSCHED_DEFAULT_TSLICE_MS    30
>  #define CSCHED_CREDITS_PER_MSEC     10
> -#define CSCHED_CREDITS_PER_TSLICE   \
> -    (CSCHED_CREDITS_PER_MSEC * CSCHED_MSECS_PER_TSLICE)
> -#define CSCHED_CREDITS_PER_ACCT     \
> -    (CSCHED_CREDITS_PER_MSEC * CSCHED_MSECS_PER_TICK * CSCHED_TICKS_PER_ACCT)
>
>
>  /*
> @@ -113,6 +107,8 @@
>  */
>  static bool_t __read_mostly sched_credit_default_yield;
>  boolean_param("sched_credit_default_yield", sched_credit_default_yield);
> +static int __read_mostly sched_credit_tslice_ms = CSCHED_DEFAULT_TSLICE_MS;
> +integer_param("sched_credit_tslice_ms", sched_credit_tslice_ms);
>
>  /*
>  * Physical CPU
> @@ -176,6 +172,9 @@ struct csched_private {
>     uint32_t credit;
>     int credit_balance;
>     uint32_t runq_sort;
> +    /* Period of master and tick in milliseconds */
> +    unsigned tslice_ms, tick_period_us, ticks_per_tslice;
> +    unsigned credits_per_tslice;
>  };
>
>  static void csched_tick(void *_cpu);
> @@ -326,7 +325,7 @@ csched_free_pdata(const struct scheduler
>
>     spin_lock_irqsave(&prv->lock, flags);
>
> -    prv->credit -= CSCHED_CREDITS_PER_ACCT;
> +    prv->credit -= prv->credits_per_tslice;
>     prv->ncpus--;
>     cpu_clear(cpu, prv->idlers);
>     cpu_clear(cpu, prv->cpus);
> @@ -360,19 +359,19 @@ csched_alloc_pdata(const struct schedule
>     spin_lock_irqsave(&prv->lock, flags);
>
>     /* Initialize/update system-wide config */
> -    prv->credit += CSCHED_CREDITS_PER_ACCT;
> +    prv->credit += prv->credits_per_tslice;
>     prv->ncpus++;
>     cpu_set(cpu, prv->cpus);
>     if ( prv->ncpus == 1 )
>     {
>         prv->master = cpu;
>         init_timer(&prv->master_ticker, csched_acct, prv, cpu);
> -        set_timer(&prv->master_ticker, NOW() +
> -                  MILLISECS(CSCHED_MSECS_PER_TICK) * CSCHED_TICKS_PER_ACCT);
> +        set_timer(&prv->master_ticker,
> +                  NOW() + MILLISECS(prv->tslice_ms));
>     }
>
>     init_timer(&spc->ticker, csched_tick, (void *)(unsigned long)cpu, cpu);
> -    set_timer(&spc->ticker, NOW() + MILLISECS(CSCHED_MSECS_PER_TICK));
> +    set_timer(&spc->ticker, NOW() + MICROSECS(prv->tick_period_us) );
>
>     INIT_LIST_HEAD(&spc->runq);
>     spc->runq_sort_last = prv->runq_sort;
> @@ -1002,7 +1001,7 @@ csched_acct(void* dummy)
>          * for one full accounting period. We allow a domain to earn more
>          * only when the system-wide credit balance is negative.
>          */
> -        credit_peak = sdom->active_vcpu_count * CSCHED_CREDITS_PER_ACCT;
> +        credit_peak = sdom->active_vcpu_count * prv->credits_per_tslice;
>         if ( prv->credit_balance < 0 )
>         {
>             credit_peak += ( ( -prv->credit_balance
> @@ -1014,7 +1013,7 @@ csched_acct(void* dummy)
>
>         if ( sdom->cap != 0U )
>         {
> -            credit_cap = ((sdom->cap * CSCHED_CREDITS_PER_ACCT) + 99) / 100;
> +            credit_cap = ((sdom->cap * prv->credits_per_tslice) + 99) / 100;
>             if ( credit_cap < credit_peak )
>                 credit_peak = credit_cap;
>
> @@ -1092,10 +1091,10 @@ csched_acct(void* dummy)
>                 }
>
>                 /* Lower bound on credits */
> -                if ( credit < -CSCHED_CREDITS_PER_TSLICE )
> +                if ( credit < -prv->credits_per_tslice )
>                 {
>                     CSCHED_STAT_CRANK(acct_min_credit);
> -                    credit = -CSCHED_CREDITS_PER_TSLICE;
> +                    credit = -prv->credits_per_tslice;
>                     atomic_set(&svc->credit, credit);
>                 }
>             }
> @@ -1117,7 +1116,7 @@ csched_acct(void* dummy)
>                 }
>
>                 /* Upper bound on credits means VCPU stops earning */
> -                if ( credit > CSCHED_CREDITS_PER_TSLICE )
> +                if ( credit > prv->credits_per_tslice )
>                 {
>                     __csched_vcpu_acct_stop_locked(prv, svc);
>                     /* Divide credits in half, so that when it starts
> @@ -1141,8 +1140,8 @@ csched_acct(void* dummy)
>     prv->runq_sort++;
>
>  out:
> -    set_timer( &prv->master_ticker, NOW() +
> -            MILLISECS(CSCHED_MSECS_PER_TICK) * CSCHED_TICKS_PER_ACCT );
> +    set_timer( &prv->master_ticker,
> +               NOW() + MILLISECS(prv->tslice_ms));
>  }
>
>  static void
> @@ -1169,7 +1168,7 @@ csched_tick(void *_cpu)
>      */
>     csched_runq_sort(prv, cpu);
>
> -    set_timer(&spc->ticker, NOW() + MILLISECS(CSCHED_MSECS_PER_TICK));
> +    set_timer(&spc->ticker, NOW() + MICROSECS(prv->tick_period_us) );
>  }
>
>  static struct csched_vcpu *
> @@ -1375,7 +1374,7 @@ csched_schedule(
>      * Return task to run next...
>      */
>     ret.time = (is_idle_vcpu(snext->vcpu) ?
> -                -1 : MILLISECS(CSCHED_MSECS_PER_TSLICE));
> +                -1 : MILLISECS(prv->tslice_ms));
>     ret.task = snext->vcpu;
>
>     CSCHED_VCPU_CHECK(ret.task);
> @@ -1469,10 +1468,9 @@ csched_dump(const struct scheduler *ops)
>            "\tweight             = %u\n"
>            "\trunq_sort          = %u\n"
>            "\tdefault-weight     = %d\n"
> -           "\tmsecs per tick     = %dms\n"
> +           "\ttslice             = %dms\n"
>            "\tcredits per msec   = %d\n"
>            "\tticks per tslice   = %d\n"
> -           "\tticks per acct     = %d\n"
>            "\tmigration delay    = %uus\n",
>            prv->ncpus,
>            prv->master,
> @@ -1481,10 +1479,9 @@ csched_dump(const struct scheduler *ops)
>            prv->weight,
>            prv->runq_sort,
>            CSCHED_DEFAULT_WEIGHT,
> -           CSCHED_MSECS_PER_TICK,
> +           prv->tslice_ms,
>            CSCHED_CREDITS_PER_MSEC,
> -           CSCHED_TICKS_PER_TSLICE,
> -           CSCHED_TICKS_PER_ACCT,
> +           prv->ticks_per_tslice,
>            vcpu_migration_delay);
>
>     cpumask_scnprintf(idlers_buf, sizeof(idlers_buf), prv->idlers);
> @@ -1526,6 +1523,13 @@ csched_init(struct scheduler *ops)
>     INIT_LIST_HEAD(&prv->active_sdom);
>     prv->master = UINT_MAX;
>
> +    prv->tslice_ms = sched_credit_tslice_ms;
> +    prv->ticks_per_tslice = CSCHED_TICKS_PER_TSLICE;
> +    if ( prv->tslice_ms < prv->ticks_per_tslice )
> +        prv->ticks_per_tslice = 1;
> +    prv->tick_period_us = prv->tslice_ms * 1000 / prv->ticks_per_tslice;
> +    prv->credits_per_tslice = CSCHED_CREDITS_PER_MSEC * prv->tslice_ms;
> +
>     return 0;
>  }
>
> @@ -1550,13 +1554,16 @@ static void csched_tick_suspend(const st
>
>  static void csched_tick_resume(const struct scheduler *ops, unsigned int cpu)
>  {
> +    struct csched_private *prv;
>     struct csched_pcpu *spc;
>     uint64_t now = NOW();
>
>     spc = CSCHED_PCPU(cpu);
>
> -    set_timer(&spc->ticker, now + MILLISECS(CSCHED_MSECS_PER_TICK)
> -            - now % MILLISECS(CSCHED_MSECS_PER_TICK) );
> +    prv = CSCHED_PRIV(ops);
> +
> +    set_timer(&spc->ticker, now + MICROSECS(prv->tick_period_us)
> +            - now % MICROSECS(prv->tick_period_us) );
>  }
>
>  static struct csched_private _csched_priv;
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-devel
>

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.