Re: [PATCH] Time sliced CFQ #2

[Jens Axboe <axboe@suse.de>]

On Mon, Dec 06 2004, Robert Love wrote:
> On Mon, 2004-12-06 at 00:00 -0500, Kyle Moffett wrote:
> 
> > What about this:
> > 
> > nice = x;		/* -20 to 20 */
> > ioprio = y;		/* -40 to 40 */
> > effective_ioprio = clamp(x+y);	/* -20 to 20 */
> > 
> > This would allow tuning processes for unusual contrasts with the ioprio 
> > call.
> > On the other hand, it would allow us to just brute force "adjust" a 
> > process with
> > the nice command in the usual way without any changes to the "nice" 
> > source.
> > 
> > I also thought of a different effective ioprio calculation that scales
> > instead of clamping:
> 
> I think the complication of all of this demonstrates the overcomplexity.
> I think we need to either
> 
> 	(1) separate the two values.  we have a scheduling
> 	    priority (distributing the finite resource of
> 	    processor time) and an I/O priority (distributing
> 	    the finite resource of disk bandwidth).
> 	(2) just have a single value.

They are inherently seperate entities, I don't think mixing them up is a
good idea. IO priorities also includes things like attempting to
guarentee disk bandwidth, it isn't always just a 'nice' value.

But lets not get carried away in a pointless discussion. The actual
setting and query interface for io priorities is by far the smallest and
least critical piece of code :-)

-- 
Jens Axboe