2006-01-28 09:59:04

by Libin Varghese

[permalink] [raw]
Subject: I/O Scheduling

Hi all,
Is there any work done on new I/O scheduling techniques (other
than as, cfq, noop, deadline)?

Regards,
Libin Varghese


2006-01-28 17:55:04

by Florian Schmidt

[permalink] [raw]
Subject: Re: I/O Scheduling

On Sat, 28 Jan 2006 15:28:54 +0530
Libin Varghese <[email protected]> wrote:

> Hi all,
> Is there any work done on new I/O scheduling techniques (other
> than as, cfq, noop, deadline)?

Hi,

i'm also interested in these. Especially I/O priorities per process/task
similar to scheduling priorities. It would be just awesome to be able to
give i.e. a hd recording program (or any other data aquisition or
playback program) a high I/O priority.

Image no buffer overruns ever in a hd recorder. Or no more video
dropouts when watching a movie and at the same time copying a file from
the same partition you play the video from.

Is there any work done in this area. I faintly remember to have read
about something like this over a year ago, but have forgotten all the
specifics.

Thanks for all infos and regards,
Florian Schmidt

--
Palimm Palimm!
http://tapas.affenbande.org

2006-01-28 18:08:43

by Lee Revell

[permalink] [raw]
Subject: Re: I/O Scheduling

On Sat, 2006-01-28 at 18:54 +0100, Florian Schmidt wrote:
>
> i'm also interested in these. Especially I/O priorities per
> process/task similar to scheduling priorities. It would be just
> awesome to be able to give i.e. a hd recording program (or any other
> data aquisition or playback program) a high I/O priority.
>

I believe it's already implemented for the CFQ scheduler only, but the
patch does not seem to be in mainline.

Jens, what's the status of this?

Lee

2006-01-28 19:17:26

by Jens Axboe

[permalink] [raw]
Subject: Re: I/O Scheduling

On Sat, Jan 28 2006, Lee Revell wrote:
> On Sat, 2006-01-28 at 18:54 +0100, Florian Schmidt wrote:
> >
> > i'm also interested in these. Especially I/O priorities per
> > process/task similar to scheduling priorities. It would be just
> > awesome to be able to give i.e. a hd recording program (or any other
> > data aquisition or playback program) a high I/O priority.
> >
>
> I believe it's already implemented for the CFQ scheduler only, but the
> patch does not seem to be in mainline.
>
> Jens, what's the status of this?

It's merged, since 2.6.13.

--
Jens Axboe

2006-01-28 19:44:41

by Lee Revell

[permalink] [raw]
Subject: Re: I/O Scheduling

On Sat, 2006-01-28 at 20:18 +0100, Jens Axboe wrote:
> On Sat, Jan 28 2006, Lee Revell wrote:
> > On Sat, 2006-01-28 at 18:54 +0100, Florian Schmidt wrote:
> > >
> > > i'm also interested in these. Especially I/O priorities per
> > > process/task similar to scheduling priorities. It would be just
> > > awesome to be able to give i.e. a hd recording program (or any other
> > > data aquisition or playback program) a high I/O priority.
> > >
> >
> > I believe it's already implemented for the CFQ scheduler only, but the
> > patch does not seem to be in mainline.
> >
> > Jens, what's the status of this?
>
> It's merged, since 2.6.13.
>

OK, I was looking at an old patch, the API must have changed. Can it be
controlled per thread, independently of the nice value/RT priority?

Lee

2006-01-28 21:28:01

by Jens Axboe

[permalink] [raw]
Subject: Re: I/O Scheduling

On Sat, Jan 28 2006, Lee Revell wrote:
> On Sat, 2006-01-28 at 20:18 +0100, Jens Axboe wrote:
> > On Sat, Jan 28 2006, Lee Revell wrote:
> > > On Sat, 2006-01-28 at 18:54 +0100, Florian Schmidt wrote:
> > > >
> > > > i'm also interested in these. Especially I/O priorities per
> > > > process/task similar to scheduling priorities. It would be just
> > > > awesome to be able to give i.e. a hd recording program (or any other
> > > > data aquisition or playback program) a high I/O priority.
> > > >
> > >
> > > I believe it's already implemented for the CFQ scheduler only, but the
> > > patch does not seem to be in mainline.
> > >
> > > Jens, what's the status of this?
> >
> > It's merged, since 2.6.13.
> >
>
> OK, I was looking at an old patch, the API must have changed. Can it be
> controlled per thread, independently of the nice value/RT priority?

Yes it can, get a recent util-linux and look at ionice (there's a man
page, too).

--
Jens Axboe