Re: + mm-balance_dirty_pages-reduce-calls-to-global_page_state-to-reduce-c ache-references.patch added to -mm tree

[Peter Zijlstra <peterz@infradead.org>]

On Wed, 2009-09-02 at 17:57 +0800, Wu Fengguang wrote:
> On Wed, Sep 02, 2009 at 04:31:40PM +0800, Peter Zijlstra wrote:
> > On Sat, 2009-08-22 at 20:11 +0200, Peter Zijlstra wrote:
> > > > > +           /* always throttle if over threshold */
> > > > > +           if (nr_reclaimable + nr_writeback < dirty_thresh) {
> > > > 
> > > > That 'if' is a big behavior change. It effectively blocks every one
> > > > and canceled Peter's proportional throttling work: the less a process
> > > > dirtied, the less it should be throttled.
> > > 
> > > Hmm, I think you're right, I had not considered that, thanks for
> > > catching that.
> > 
> > So in retrospect I think I might have been wrong here.
> > 
> > The per task thing causes the bdi limit to be lower than the bdi limit
> > based on writeback speed alone. That is, the more a task dirties, the
> > lower the bdi limit is as seen for that task.
> 
> Right. If I understand it right, there will be a safety margin of about
> (1/8) * dirty_limit for 1 heavy dirtier case, and that gap scales down
> when there are more concurrent heavy dirtiers.

Right, with say 4 heavy writers the gap will be 1/4-th of 1/8-th, which
is 1/32-nd.

With the side node that I think 1/8 is too much on large memory systems,
and I have posted a sqrt patch numerous times, but I don't think we've
ever found out if that helps or not...

> In principle, the ceiling will be a bit higher for a light dirtier to
> make it easy to pass in the presence of more heavy dirtiers.

Right.

> > So if we get a task that generates tons of dirty pages (dd) then it
> > won't ever actually hit the full dirty limit, even if its the only task
> > on the system, and this outer if() will always be true.
> 
> Right, we have the safety margin :)
> 
> > Only when we actually saturate the full dirty limit will we fall through
> > and throttle, but that is ok -- we want to enforce the full limit.
> > 
> > In short, a very aggressive dirtier will have a bdi limit lower than the
> > total limit (at all times) leaving a little room at the top for the
> > occasional dirtier to make quick progress.
> > 
> > Wu, does that cover the scenario you had in mind?
> 
> Yes thanks! Please correct me if wrong:
> - the lower-ceiling-for-heavier-dirtier algorithm in task_dirty_limit()
>   is elegant enough to prevent heavy dirtier to block light ones

ack

> - the test (nr_reclaimable + nr_writeback < dirty_thresh) is not
>   relevant in normal, but can be kept for safety in the form of
> 
>           if (bdi_nr_reclaimable + bdi_nr_writeback < bdi_thresh &&
>               nr_reclaimable + nr_writeback < dirty_thresh)
>                   break;

ack

> - clip_bdi_dirty_limit() could be removed: we have been secured by the
>   above test

ack.