Re: [RFC][PATCH v5 00/14] sched: packing tasks

[Catalin Marinas]

On Wed, Nov 13, 2013 at 04:13:57PM +0000, Arjan van de Ven wrote:
> On 11/12/2013 3:14 PM, Catalin Marinas wrote:
> > On 12 Nov 2013, at 16:48, Arjan van de Ven <arjan@xxxxxxxxxxxxxxx> wrote:
> >> On 11/11/2013 10:18 AM, Catalin Marinas wrote:
> >>> The ordering is based on the actual C-state, so a simple way is to wake
> >>> up the CPU in the shallowest C-state. With asymmetric configurations
> >>> (big.LITTLE) we have different costs for the same C-state, so this would
> >>> come in handy.
> >>
> >> btw I was considering something else; in practice CPUs will be in the deepest state..
> >> ... at which point I was going to go with some other metrics of what is best from a platform level
> >
> > I agree, other metrics are needed.  The problem is that we currently
> > only have (relatively, guessed from the target residency) the cost of
> > transition from a C-state to a P-state (for the latter, not sure which).
> > But we donât know what the power (saving) on that C-state is nor the one
> > at a P-state (and vendors reluctant to provide such information). So the
> > best the scheduler can do is optimise the wake-up cost and blindly assume
> > that deeper C-state on a CPU is more efficient than lower P-states on two
> > other CPUs (or the other way around).
> 
> for picking the cpu to wake on there are also low level physical kind of things
> we'd want to take into account on the intel side.

Are these static and could they be hidden behind some cost number in a
topology description? If they are dynamic, we would need arch or driver
hooks to give some cost or priority number that the scheduler can use.

-- 
Catalin
--
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/