RE: bug in sched.c:activate_task()

["Chen, Kenneth W" <kenneth.w.chen@intel.com>]

On Tue, 5 Oct 2004, Con Kolivas wrote:
> We used to compare jiffy difference in can_migrate_task by comparing it
> to cache_decay_ticks. Somewhere in the merging of sched_domains it was
> changed to task_hot which uses timestamp.


On Tuesday, October 05, 2004 12:10 AM, Ingo Molnar wrote:
> yep, that's fishy. Kenneth, could you try the simple patch below? It gets
> rid of task_hot() in essence. If this works out we could try it - it gets
> rid of some more code from sched.c too. Perhaps SD_WAKE_AFFINE is enough
> control.
>
> --- kernel/sched.c.orig	2004-10-05 08:28:42.295395160 +0200
> +++ kernel/sched.c	2004-10-05 09:07:44.081389576 +0200
> @@ -180,7 +180,7 @@ static unsigned int task_timeslice(task_
>  	else
>  		return SCALE_PRIO(DEF_TIMESLICE, p->static_prio);
>  }
> -#define task_hot(p, now, sd) ((now) - (p)->timestamp < (sd)->cache_hot_time)
> +#define task_hot(p, now, sd) 0
>
>  enum idle_type
>  {

We have experimented with similar thing, via bumping up sd->cache_hot_time to
a very large number, like 1 sec.  What we measured was a equally low throughput.
But that was because of not enough load balancing, we are seeing is large amount
of idle time.

- Ken