[PATCH 09/14] sched: evaluate the activity level of the system

[Vincent Guittot]

On 22 May 2013 18:50, Morten Rasmussen <morten.rasmussen at arm.com> wrote:
> On Thu, Apr 25, 2013 at 06:23:25PM +0100, Vincent Guittot wrote:
>> We evaluate the activity level of CPUs, groups and domains in order to define
>> how many CPUs are required to handle the current activity.
>>
>> Signed-off-by: Vincent Guittot <vincent.guittot at linaro.org>
>> ---
>>  kernel/sched/fair.c |   19 +++++++++++++++++++
>>  1 file changed, 19 insertions(+)
>>
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index 954adfd..234ecdd 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -3465,7 +3465,22 @@ static int check_pack_buddy(int cpu, struct task_struct *p)
>>               return true;
>>
>>       return false;
>> +}
>> +
>> +static int get_cpu_activity(int cpu)
>> +{
>> +     struct rq *rq = cpu_rq(cpu);
>> +     u32 sum = rq->avg.runnable_avg_sum;
>> +     u32 period = rq->avg.runnable_avg_period;
>> +
>> +     sum = min(sum, period);
>> +
>> +     if (sum == period) {
>> +             u32 overload = rq->nr_running > 1 ? 1 : 0;
>> +             return power_of(cpu) + overload;
>> +     }
>
> Adding one to indicate cpu overload seems to work well in the cases that
> I can think of. It may take some time to have effect if the overloaded
> cpu is in a group with nearly overloaded cpus. Adding one won't enable
> more cpus in another group until they have all reached overload.

This output expresses the activity of the CPU and i don't want to hack
it unnecessarily.  The +1 is only used to make difference between a
cpu fully loaded by 1 task and a cpu which need help of other cpus
because it is fully loaded by several tasks which share the runtime of
 the cpu.

When we define how many cpus are needed for the packing effort, we can
take some margins like the is_buddy_full function.

>
> Reaching sum == period takes quite a long time. Would it be an idea to
> use sum*1024 >= period*1000 like you do in is_buddy_full()?

We could probably use this margin when we update the list of cpus that
are involved in the packing effort

Thanks for the comments
Vincent

>
> I think that would be more consistent since overloaded is essential the
> same as full.
>
> Cheers,
> Morten
>
>>
>> +     return (sum * power_of(cpu)) / period;
>>  }
>>
>>  /*
>> @@ -4355,6 +4370,7 @@ struct sd_lb_stats {
>>       struct sched_group *busiest; /* Busiest group in this sd */
>>       struct sched_group *this;  /* Local group in this sd */
>>       unsigned long total_load;  /* Total load of all groups in sd */
>> +     unsigned long total_activity;  /* Total activity of all groups in sd */
>>       unsigned long total_pwr;   /*   Total power of all groups in sd */
>>       unsigned long avg_load;    /* Average load across all groups in sd */
>>
>> @@ -4383,6 +4399,7 @@ struct sd_lb_stats {
>>  struct sg_lb_stats {
>>       unsigned long avg_load; /*Avg load across the CPUs of the group */
>>       unsigned long group_load; /* Total load over the CPUs of the group */
>> +     unsigned long group_activity; /* Total activity of the group */
>>       unsigned long sum_nr_running; /* Nr tasks running in the group */
>>       unsigned long sum_weighted_load; /* Weighted load of group's tasks */
>>       unsigned long group_capacity;
>> @@ -4629,6 +4646,7 @@ static inline void update_sg_lb_stats(struct lb_env *env,
>>               }
>>
>>               sgs->group_load += load;
>> +             sgs->group_activity += get_cpu_activity(i);
>>               sgs->sum_nr_running += nr_running;
>>               sgs->sum_weighted_load += weighted_cpuload(i);
>>               if (idle_cpu(i))
>> @@ -4752,6 +4770,7 @@ static inline void update_sd_lb_stats(struct lb_env *env,
>>                       return;
>>
>>               sds->total_load += sgs.group_load;
>> +             sds->total_activity += sgs.group_activity;
>>               sds->total_pwr += sg->sgp->power;
>>
>>               /*
>> --
>> 1.7.9.5
>>
>>
>