Re: [PATCH/RFC] slab: handle memoryless nodes efficiently

[Lee Schermerhorn <Lee.Schermerhorn@hp.com>]

On Thu, 2009-10-29 at 17:30 -0400, Christoph Lameter wrote:
> Maybe better introduce an alternative to numa_node_id that refers to the
> next memory node?
> 
> numa_mem_node_id?
> 
> We can then use that in various subsystems and could use it consistently
> also in slab.c

Where should we put it?  In page_alloc.c that manages the zonelists.

> 
> One problem with such a scheme (and also this patch) is that multiple
> memory nodes may be at the same distance to a processor on a memoryless
> node. Should the allocation not take memory from any of these nodes?

Well, this is the case for normal page allocations as well, but we
choose one, in build_zonelists(), that we'll use whenever a page
allocation overflows the target node selected by the mempolicy.  So,
that seemed a reasonable node to use for slab allocations.  

Thoughts?

Lee

--
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:"dont@kvack.org"> email@kvack.org </a>