Date: Tue, 14 Jul 2009 16:40:08 +0200
From: Jan Kara <jack@suse.cz>
To: Chris Mason <chris.mason@oracle.com>, Jan Kara <jack@suse.cz>,
       Jens Axboe <jens.axboe@oracle.com>, linux-kernel@vger.kernel.org,
       rjw@sisk.pl
Subject: Re: [PATCH] Fix congestion_wait() sync/async vs read/write
	confusion
Message-ID: <20090714144008.GC17633@duck.suse.cz>
References: <20090708184703.GW23611@kernel.dk> <20090708191238.GE18008@think> <20090714114419.GA17633@duck.suse.cz> <20090714131215.GF3783@think>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20090714131215.GF3783@think>
User-Agent: Mutt/1.5.17 (2007-11-01)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2594
Lines: 58

On Tue 14-07-09 09:12:15, Chris Mason wrote:
> On Tue, Jul 14, 2009 at 01:44:19PM +0200, Jan Kara wrote:
> > On Wed 08-07-09 15:12:38, Chris Mason wrote:
> > > On Wed, Jul 08, 2009 at 08:47:03PM +0200, Jens Axboe wrote:
> > > > Hi,
> > > > 
> > > > This one isn't great, we currently have broken congestion wait logic in
> > > > the kernel. 2.6.30 is impacted as well, so this patch should go to
> > > > stable too once it's in -git. I'll let this one simmer until tomorrow,
> > > > then ask Linus to pull it. The offending commit breaking this is
> > > > 1faa16d22877f4839bd433547d770c676d1d964c.
> > > > 
> > > > Meanwhile, it could potentially cause buffered writeout slowdowns in the
> > > > kernel. Perhaps the 2.6.30 regression in that area is caused by this?
> > > > Would be interesting if the submitter could test. I can't find the list,
> > > > CC'ing Rafael.
> > > 
> > > Even if this does slow down some workloads, the bug is not in using the
> > > correct flag ;) So, I'd ack this one.
> > > 
> > > Jan Kara was able to reproduce the tiobench 2.6.30 regression, so I've
> > > cc'd him and kept the patch below.
> >   Thanks for the patch Chris. I've remeasured tiobench with the 2.6.30 +
> > the fix but it didn't help (which is not too surprising as what I observe
> > is most likely CFQ related as there's no regression with NOOP scheduler).
> >   Just to recall:
> >     2.6.29 (CFQ)           Avg    StdDev
> > 8   38.01 40.26 39.69 ->  39.32  0.955092
> > 16  40.09 38.18 40.05 ->  39.44  0.891104
> > 
> >     2.6.30-rc8 (CFQ)
> > 8   36.67 36.81 38.20 ->  37.23  0.69062
> > 16  37.45 36.47 37.46 ->  37.13  0.464351
> > 
> >     2.6.30-rc8+fix (CFQ)
> > 8   37.56 37.38 37.98 ->  37.64  0.251396
> > 16  38.11 36.71 37.18 ->  37.33  0.581741
> > 
> >   So with the fix there's no statistically significant difference and we
> > are still below 2.6.29 results. I'm now going to retest with the WRITE_SYNC
> > changes reverted.
> 
> Well, its good the patch didn't make things worse ;)  I didn't have the
> highest hopes that it would resolve the regression, but thanks for
> testing!
  I've now tried to revert everything which looked WRITE_SYNC related but
it didn't help either. Now, I'm trying to basically bisect CFQ changes and
I'll see whether it goes somewhere...

									Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/