Re: [patch 0/4] ide: Break ide_lock to per-hwgroup lock

From: Jens Axboe
Date: Tue Sep 27 2005 - 10:27:13 EST


On Tue, Sep 27 2005, Ravikiran G Thirumalai wrote:
> On Tue, Sep 27, 2005 at 03:36:40PM +0200, Bartlomiej Zolnierkiewicz wrote:
> > On 9/7/05, Jens Axboe <axboe@xxxxxxx> wrote:
> > > On Wed, Sep 07 2005, Ravikiran G Thirumalai wrote:
> > > > On Wed, Sep 07, 2005 at 11:19:24AM +0200, Jens Axboe wrote:
> > > > > On Tue, Sep 06 2005, Ravikiran G Thirumalai wrote:
> > > > > > The following patchset breaks down the global ide_lock to per-hwgroup lock.
> > > > > > We have taken the following approach.
> > > > >
> > > > > Curious, what is the point of this?
> > > > >
> > > >
> > >
> > > I'm asking because I've never heard anyone complain about IDE lock
> > > contention and a proper patch usually comes with analysis of why it is
> > > needed.
> >
> > Since ide_lock spinlock is used for all drives as queue lock and for all
> > controllers as IDE lock I guess that with multiple controllers there is a lot
> > contention on it...
> >
> > Breaking ide_lock is fine with me however seeing numbers would
> > greatly help in getting wider acceptance for this change, Ravikiran?
> >
>
> With the lock breakdown, Iozone read performance went up by 5.5% on a 2 way
> x86 xeon box, with two hwifs and two processes reading from disks connected
> to each hwif. There seems to be a small regression on Iozone write
> tests -- about 1.5%. Maybe this is measurement noise as there is no apparent
> reason for this. We plan to run more tests to see if the regression for
> writes is statistically significant.

You should run it eg 10 times on each kernel to get a feel for the
variance of the results. Were you testing 2 or 4 disks?

Some profile information for both kernels would also be paramount.

> Btw, there was some problem with the moving of tuning code in this patchset
> causing disks not to work in DMA mode. We were fixing that; will publish a
> newer patchset with test results soon.

I take it the numbers posted were for DMA enabled on all disks?

--
Jens Axboe

-
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/