[RFC PATCH v3 3/6] sched: pack small tasks

[alex.shi@intel.com (Alex Shi)]

On 03/27/2013 12:33 PM, Preeti U Murthy wrote:
> Hi Peter,
> 
> On 03/26/2013 06:07 PM, Peter Zijlstra wrote:
>> On Fri, 2013-03-22 at 13:25 +0100, Vincent Guittot wrote:
>>> +static bool is_light_task(struct task_struct *p)
>>> +{
>>> +       /* A light task runs less than 20% in average */
>>> +       return ((p->se.avg.runnable_avg_sum  * 5) <
>>> +                       (p->se.avg.runnable_avg_period));
>>> +}
>>
>> OK, so we have a 'problem' here, we initialize runnable_avg_* to 0, but
>> we want to 'assume' a fresh task is fully 'loaded'. IIRC Alex ran into
>> this as well.
>>
>> PJT, do you have any sane solution for this, I forgot what the result
>> of the last discussion was -- was there any?
> 
> The conclusion after last discussion between PJT and Alex was that the
> load contribution of a fresh task be set to "full" during "__sched_fork()".
> 
> task->se.avg.load_avg_contrib = task->se.load.weight during
> __sched_fork() is reflected in the latest power aware scheduler patchset
> by Alex.

Yes, the new forked runnable load was set as full utilisation in V5
power aware scheduling. PJT, Mike and I both agree on this. PJT just
discussion how to give the full load to new forked task. and we get
agreement in my coming V6 power aware scheduling patchset.

> 
> Thanks
> 
> Regards
> Preeti U Murthy
>>
> 


-- 
Thanks Alex