Re: [PATCH][5/?] count writeback pages in nr_scanned

[Andrew Morton <akpm@osdl.org>]

Marcelo Tosatti <marcelo.tosatti@cyclades.com> wrote:
>
> On Wed, Jan 05, 2005 at 01:50:51PM -0500, Rik van Riel wrote:
> > On Wed, 5 Jan 2005, Andrea Arcangeli wrote:
> > 
> > >Another unrelated problem I have in this same area and that can explain
> > >VM troubles at least theoretically, is that blk_congestion_wait is
> > >broken by design. First we cannot wait on random I/O not related to
> > >write back. Second blk_congestion_wait gets trivially fooled by
> > >direct-io for example. Plus the timeout may cause it to return too early
> > >with slow blkdev.

That's true, as we discussed a couple of months back.  But the current code
is nice and simple and has been there for a couple of years with no
observed problems.


> > Or the IO that just finished, finished for pages in
> > another memory zone, or pages we won't scan again in
> > our current go-around through the VM...
> 
> Thing is there is no distinction between pages which have been written out 
> for what purpose at the block level. 
> 
> One can conjecture the following: per-zone waitqueue to be awakened from 
> end_page_writeback() (for PG_reclaim pages only of course), and a function
> to wait on the perzone waitqueue:
> 
>  wait_vm_writeback (zone, timeout);
> 
> Instead of the current blk_congestion_wait() on try_to_free_pages/balance_pgdat.
> 

The caller would need to wait on all the zones which can satisfy the
caller's allocation request.  A bit messy, although not rocket science. 
One would have to be careful to avoid additional CPU consumption due to
delivery of multiple wakeups at each I/O completion.

We should be able to demonstrate that such a change really fixes some
problem though.  Otherwise, why bother?