On Tue, 21 Aug 2007, Dave McCracken wrote:

quoted text
> On Monday 20 August 2007, Christoph Lameter wrote:
> > 1. First reclaiming non dirty pages. Dirty pages are deferred until rec=
laim
quoted text
> > =A0 =A0has reestablished the high marks. Then all the dirty pages (the =
laundry)
quoted text
> > =A0 =A0is written out.
>=20
> I don't buy it.  What happens when there aren't enough clean pages in the=
=20
quoted text
> system to achieve the high water mark?  I'm guessing we'd get a quick OOM=
 (as=20
quoted text
> observed by Peter).

We reclaim the clean pages that there are (removing the executable=20
pages from memory) and then we do writeback.

The quick OOM is due to throttling not working right AFAIK.

quoted text
> > 2. Reclaim is essentially complete during the writeout phase. So we rem=
ove
quoted text
> > =A0 =A0PF_MEMALLOC and allow recursive reclaim if we still run into tro=
uble
quoted text
> > =A0 =A0during writeout.
>=20
> You're assuming the system is static and won't allocate new pages behind =
your=20
quoted text
> back.  We could be back to critically low memory before the write happens=
=2E

Yes and that occurs now too.

quoted text
> More broadly, we need to be proactive about getting dirty pages cleaned b=
efore=20
quoted text
> they consume the system.  Deferring the write just makes it harder to kee=
p=20
quoted text
> up.

Cleaning dirty pages through writeout consumes memory. Writing dirty pages=
=20
out early makes the memory situation even worse.