Re: [RFC 2.6.11-rc2-mm2 7/7] mm: manual page migration -- sys_page_migrate

["Martin J. Bligh" <mbligh@aracnet.com>]

--On Wednesday, February 16, 2005 01:44:01 +0100 Andi Kleen <ak@suse.de> wrote:

>> SGI had code in IRIX to do that kind of thing (automatically move a page to
>> the node where most of the references were coming from).  Never worked very 
>> well, I have been told.   So our bias is away from such "automatic" page
>> migration schemes and toward "manual" methods driven either by a user
>> command or a user-level program such as a batch scheduler.
> 
> I tried something similar too (scheduling a task to the node with most of 
> its memory) and it also never worked very well.

I'm talking about doing it the other way around though - just allocating
the memory local to the task, not bringing the task to the memory.

If kswapd would keep up, we'd be OK - it should keep enough memory free
on the local node that we'd always get local free pages for the new bits.
If we go into fallback and immediate reclaim from __alloc_pages, then I
think everything starts to fall apart.

Perhaps we should just make sure most of the reclaim is being done from
kswapd - we're trying to work out exactly how much of the reclaim happens
where right at the moment.

M.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"aart@kvack.org"> aart@kvack.org </a>