Re: [PATCH] slob: poor man's NUMA, take 2.

[Nick Piggin <nickpiggin@yahoo.com.au>]

Paul Mundt wrote:
> Here's an updated copy of the patch adding simple NUMA support to SLOB,
> against the current -mm version of SLOB this time.
> 
> I've tried to address all of the comments on the initial version so far,
> but there's obviously still room for improvement.
> 
> This approach is not terribly scalable in that we still end up using a
> global freelist (and a global spinlock!) across all nodes, making the
> partial free page lookup rather expensive. The next step after this will
> be moving towards split freelists with finer grained locking.

I just think that this is not really a good intermediate step because
you only get NUMA awareness from the first allocation out of a page. I
guess that's an easy no-brainer for bigblock allocations, but for SLUB
proper, it seems not so good.

For a lot of workloads you will have a steady state where allocation and
freeing rates match pretty well and there won't be much movement of pages
in and out of the allocator. In this case it will be back to random
allocations, won't it?

-- 
SUSE Labs, Novell Inc.

--
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>