Re: [RFC] Simple Slab: A slab allocator with minimal meta information
From: Matt Mackall
Date: Wed Aug 09 2006 - 22:06:29 EST
On Wed, Aug 09, 2006 at 05:52:00PM -0700, Christoph Lameter wrote:
> This is by no means complete and probably full of bugs. Feedback and help
> wanted! I have tried to switch over two minor system caches (memory
> policies) to use simple slab and it seems to work. We probably have some
> way to go before we could do performance tests.
There's probably enough here to shim in a regular slab and kmalloc
interface layer to run a desktop machine.
> The Simple Slab is not NUMA capable at this point and I think the
> NUMAness may better be implemented in a different way. Maybe we
> could understand the Simple Slab as a lower layer and then add all
> the bells and whistles including NUMAness, proc API, kmalloc caches
> etc. on top as a management layer for this lower level
I think a layered approach to handling NUMA and the like makes an
awful lot of sense here. And probably greatly simplifies locking, etc.
Also, I like that you've gone to off-slab accounting. Not only does
this simplify things overall, it's good for memory footprint and
possibly better on cache footprint.
It's gonna need a better name though..
Mathematics is the supreme nostalgia of our time.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/