Re: [PATCH 3/3] writeback: Do not congestion sleep when there areno congested BDIs

[Mel Gorman]

On Fri, Aug 27, 2010 at 09:11:06AM +0800, Wu Fengguang wrote:
> On Fri, Aug 27, 2010 at 04:23:24AM +0800, Mel Gorman wrote:
> > On Thu, Aug 26, 2010 at 08:17:35PM +0200, Johannes Weiner wrote:
> > > On Thu, Aug 26, 2010 at 06:42:45PM +0100, Mel Gorman wrote:
> > > > On Fri, Aug 27, 2010 at 02:38:43AM +0900, Minchan Kim wrote:
> > > > > On Thu, Aug 26, 2010 at 04:14:16PM +0100, Mel Gorman wrote:
> > > > > > If congestion_wait() is called with no BDIs congested, the caller will
> > > > > > sleep for the full timeout and this is an unnecessary sleep. This patch
> > > > > > checks if there are BDIs congested. If so, it goes to sleep as normal.
> > > > > > If not, it calls cond_resched() to ensure the caller is not hogging the
> > > > > > CPU longer than its quota but otherwise will not sleep.
> > > > > > 
> > > > > > This is aimed at reducing some of the major desktop stalls reported during
> > > > > > IO. For example, while kswapd is operating, it calls congestion_wait()
> > > > > > but it could just have been reclaiming clean page cache pages with no
> > > > > > congestion. Without this patch, it would sleep for a full timeout but after
> > > > > > this patch, it'll just call schedule() if it has been on the CPU too long.
> > > > > > Similar logic applies to direct reclaimers that are not making enough
> > > > > > progress.
> > > > > > 
> > > > > > Signed-off-by: Mel Gorman <mel@xxxxxxxxx>
> > > > > > ---
> > > > > >  mm/backing-dev.c |   20 ++++++++++++++------
> > > > > >  1 files changed, 14 insertions(+), 6 deletions(-)
> > > > > > 
> > > > > > diff --git a/mm/backing-dev.c b/mm/backing-dev.c
> > > > > > index a49167f..6abe860 100644
> > > > > > --- a/mm/backing-dev.c
> > > > > > +++ b/mm/backing-dev.c
> > > > > 
> > > > > Function's decripton should be changed since we don't wait next write any more. 
> > > > > 
> > > > 
> > > > My bad. I need to check that "next write" thing. It doesn't appear to be
> > > > happening but maybe that side of things just broke somewhere in the
> > > > distant past. I lack context of how this is meant to work so maybe
> > > > someone will educate me.
> > > 
> > > On every retired io request the congestion state on the bdi is checked
> > > and the congestion waitqueue woken up.
> > > 
> > > So without congestion, we still only wait until the next write
> > > retires, but without any IO, we sleep the full timeout.
> > > 
> > > Check __freed_requests() in block/blk-core.c.
> > > 
> > 
> > Seems reasonable. Still, if there is no write IO going on and no
> > congestion there seems to be no point going to sleep for the full
> > timeout. It still feels wrong.
> 
> Yeah the stupid sleeping feels wrong. However there are ~20
> congestion_wait() callers spread randomly in VM, FS and block drivers.
> Many of them may be added by rule of thumb, however what if some of
> them happen to depend on the old stupid sleeping behavior? Obviously
> you've done extensive tests on the page reclaim paths, however that's
> far from enough to cover the wider changes made by this patch.
> 
> We may have to do the conversions case by case. Converting to
> congestion_wait_check() (see http://lkml.org/lkml/2010/8/18/292) or
> other waiting schemes.
> 

I am taking this direction now.

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab
--
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/