Re: IO queuing and complete affinity with threads (was Re: [PATCH 0/8] IO queuing and complete affinity)

[Nick Piggin]

On Fri, Feb 08, 2008 at 09:24:22AM +0100, Jens Axboe wrote:
> On Fri, Feb 08 2008, Nick Piggin wrote:
> > On Fri, Feb 08, 2008 at 08:59:55AM +0100, Jens Axboe wrote:
> > > On Fri, Feb 08 2008, Nick Piggin wrote:
> > > > And if you don't?
> > > 
> > > Well if you don't ask for anything, you wont get anything :-)
> > > As I mentioned, the patch is a playing ground for trying various setups.
> > > Everything defaults to 'do as usual', set options to setup certain test
> > > scenarios.
> > 
> > I mean if you don't know the completing CPU.
> 
> I still don't know quite what part of that patch you are referring to
> here. If you don't have queue_affinity set, queueing a new request with
> the hardware is generally done on the same CPU that just completed a
> request. That is true even without any patches.

Generally, but I guess not always. The database workloads in question
(which you might know very well about ;)) apparently has a lot of
queue empty and unplug conditions. Which I guess is the reason for
Intel's initial patch.

--
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/