Re: NUMA, page color, and the page daemon

[David Laight <david%l8s.co.uk@localhost>]

On Thu, Jun 02, 2011 at 06:55:54PM -0700, Matt Thomas wrote:
> 
> The page coloring support I've been working with on the matt-nb5-mips64
> branch has been working pretty well, except for one thing.
...
> The solution to this is to treat each color as its own collection of
> pages which has its own free target, page queues, etc.
> The page daemon would work on each page collection.
> This would mean the page daemon would need to keep a copy of its state for 
> each color.
> 
> NUMA architectures would need similar support since each memory region
> would need to serviced by its owner (for best efficiency).
...
Would this also help systems like amd64 where memory below 4G may run out?
(Not no mention the 16M limit ...)

        David

-- 
David Laight: david%l8s.co.uk@localhost