[dm-devel] high iowait problem(Bug 12309 on bugzilla.kernel.org)

[Andrew Morton]

On Thu, 15 Apr 2010 00:58:05 -0500
Steven B <damentz at gmail.com> wrote:

> Hi Andrew,
> 
> > So something between 2.6.32.3 and 2.6.32.8 inclusive rebroke the
> > kernel. __That's useful info.
> >
> > Looking at the changelogs, the only thing I can immediately see in the
> > block-core area is
> >
> > : commit cb723ba5d03bf719dbc7409b4d67572d4472ef8b
> > : Author: Dmitry Monakhov <dmonakhov at openvz.org>
> > : Date: __ Wed Jan 27 22:44:36 2010 +0300
> > :
> > : __ __ block: fix bio_add_page for non trivial merge_bvec_fn case
> >
> > which seems unlikely.
> >
> > I'm seeing just one device-mapper change whcih went in over that
> > timeframe:
> >
> > : commit e0f5cfa7c18b411634e73923841eccd3a4c0ce7f
> > : Author: Martin K. Petersen <martin.petersen at oracle.com>
> > : Date: __ Mon Jan 11 03:21:50 2010 -0500
> > :
> > : __ __ DM: Fix device mapper topology stacking
> >
> >
> 
> Although I can't provide any big details, I went through the commits between
> 2.6.32.2 and 2.6.32.8 and as many things that could affect IO and reverted
> them from my 2.6.33.2 kernel.
> 
> Here's my list of commits I reverted that seemed to improve performance locally
> running a 2.6.33.2 kernel.
> 
> cb723ba5d03bf719dbc7409b4d67572d4472ef8b (this one you suspected)
> block: fix bio_add_page for non trivial merge_bvec_fn case
> 
> b2ea8cb9c8f1937cb80b9beb50548a05bfc37819
> vmscan: do not evict inactive pages when skipping an active list scan
> 
> e4dd8ca5be771aa706b0797b3284328bb595bfa1
> fasync: split 'fasync_helper()' into separate add/remove functions
> 
> Though, I was hoping you guys could identify which one of these is
> the real culprit.

Can you go through and revert them individually please?