Re: [PATCH 1/3] CPUMETER: add cpumeter framework to the CPUSETS

From: KUROSAWA Takahiro
Date: Wed Sep 28 2005 - 05:33:43 EST


On Wed, 28 Sep 2005 19:08:22 +0900 (JST)
Hirokazu Takahashi <taka@xxxxxxxxxxxxx> wrote:

> > - The controller only controls the time_slice value and doesn't
> > care the balance of cpus (the controller leaves balancing to
> > the existing balancing code in the scheduler). So far I don't
> > have any good idea to solve this.
> Is there any problem to enhance load_balance()? Applying some weight
> to each process, it would be used for calculating load of each runqueue.

Probably we can solve this issue by applying some weight to each tasks
and calculating the load of runqueues based on that weight.
But I can not get to the implementation out of your idea so far.

> > - The controller holds a spinlock once per 1 second. I don't
> > know this is permissive or not, but the current scheduler
> > doesn't hold any spinlocks normally, so...
> I think each runqueue has its own spinlock, which is held quite often.

Ah, thanks for pointing out my mistake. Maybe I confused that with
something else.
Then it should be more permissive than I had thought, at least :)

--
KUROSAWA, Takahiro
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/