Re: [PATCH] Use MPOL_INTERLEAVE for tmpfs files

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

>> Matt has volunteered to write the mount option for this, so let's hold
>> off for a couple of days until that's done.
> 
> I had the time to do this myself.  Updated patch attached below.
> 
> Description:
> 
> This patch creates a tmpfs mount option and adds code which causes
> tmpfs allocations to be interleaved across the nodes of a NUMA machine.
> This is useful in situations where a large tmpfs file would otherwise
> consume most of the memory on a single node, forcing tasks running on
> that node to perform off-node allocations for other (i.e. non-tmpfs)
> memory needs.
> 
> Tightly synchronized HPC applications with large (on the order of
> a single node's total RAM) per-task memory requirements are an example
> of a type of application which benefits from this change.  Other
> types of workloads may prefer local tmpfs allocations, thus a mount
> option is provided.
> 
> The new mount option is "interleave=", and defaults to 0.  Any non-zero
> setting turns on interleaving.  Interleaving behavior can be changed
> via a remount operation.

Looks good to me, though I don't really know that area of code very well.
Hopefully Hugh or someone can give it a better review.

Thanks for doing that,

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>