Re: [PATCH 11/39] autonuma: CPU follow memory algorithm

[Andrea Arcangeli]

Hi,

On Wed, Mar 28, 2012 at 01:26:08PM +0200, Peter Zijlstra wrote:
> Right, so can we agree that the only case where they diverge is single
> processes that have multiple threads and are bigger than a single node (either
> in memory, cputime or both)?

I think it vastly diverges for processes that are smaller than one
node too. 1) your numa/sched goes blind with an almost arbitrary home
node, 2) your migrate-on-fault will be unable to provide an efficient
and steady async background migration.

> I've asked you several times why you care about that one case so much, but
> without answer.

If this case wasn't important to you, you wouldn't need to introduce
your syscalls.

> I'll grant you that unmodified such processes might do better with your
> stuff, however:
> 
>  - your stuff assumes there is a fair amount of locality to exploit.
> 
>    I'm not seeing how this is true in general, since data partitioning is hard
>    and for those problems where its possible people tend to already do so,
>    yielding natural points to add the syscalls.

Later, I plan to detect this and layout interleaved pages
automatically so you don't even need to manually set MPOL_INTERLEAVE.

>  - your stuff doesn't actually nest, since a guest kernel has no clue as to
>    what constitutes a node (or if there even is such a thing) it will randomly
>    move tasks around on the vcpus, with complete disrespect for whatever host
>    vcpu<->page mappings you set up.
> 
>    guest kernels actively scramble whatever relations you're building by
>    scanning, destroying whatever (temporal) locality you think you might
>    have found.

This shall work fine, running AutoNUMA in guest and host. qemu just
need to create a vtopology for the guest that matches the hardware
topology. Hard binds in the guest will also work great (they create
node locality too).

A paravirt layer could also hint the host on the vcpu switches to
shift the host numa stats across but I didn't thought too much on this
possible paravirt numa-sched optimization, it's not mandatory, just an idea.

> Related to this is that all applications that currently use mbind() and
> sched_setaffinity() are trivial to convert.

Too bad firefox isn't using mbind yet. My primary target are the 99%
of apps out there running on a 24way 2 node system or equivalent and
KVM.

I agree converting qemu to the syscalls would be trivial though.
--
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/