Messages in this thread | | | Date | Wed, 18 Mar 2009 15:07:24 +0000 | From | Mel Gorman <> | Subject | Re: [PATCH 00/35] Cleanup and optimise the page allocator V3 |
| |
On Mon, Mar 16, 2009 at 06:05:51PM +0100, Nick Piggin wrote: > On Mon, Mar 16, 2009 at 04:56:28PM +0000, Mel Gorman wrote: > > On Mon, Mar 16, 2009 at 04:53:42PM +0100, Nick Piggin wrote: > > > So yes definitely I think there should be a very real impact on > > > higher order coalescing no matter what you do. > > > > > > > Because this is not straight-forward at all, I'll put lazy buddy onto > > the back-burner and exhaust all other possibilities before revisiting it > > again. > > If it is such a big improvement, I expect *most* people will want > it and we probably should do it.
I'll be reinvestigating it in isolation. It's possible that high-order and compound pages on the PCP lists is enough of a delayed buddy merging that the benefit from lazy buddy is marginal.
> But just that it will not play > nicely with fragmentation and so you'd need to look into it and > devise some way those users can tune it to be nicer. >
Which is why I'm going to postpone it for now.
> > > unmovable zone fragmentation is more important point because it > > > eventually can destroy the movable zone. > > > > > > > Which is why rmqueue_fallback() also merges up all buddies before making > > any decisions but I accept your points. > > Right, that merge of buddies will only be able to look at what is > currently free. Wheras non-lazy buddy can pull out higher orders > before reallocating them. >
-- Mel Gorman Part-time Phd Student Linux Technology Center University of Limerick IBM Dublin Software Lab
| |