[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] PV-vNUMA issue: topology is misinterpreted by the guest

On 07/17/2015 03:27 AM, Dario Faggioli wrote:
On Fri, 2015-07-17 at 07:09 +0100, Jan Beulich wrote:
On 16.07.15 at 18:59, <boris.ostrovsky@xxxxxxxxxx> wrote:
And in general (both for PV and HVM) --- is there any reason to expose
CPU topology at all? I can see it being useful if VCPUs are pinned but
if they are not then it can make performance worse.

Indeed indeed. :-)

And in fact, this is even independent from vNUMA. Yet, I remember we
were discussing about this since the beginning of vNUMA work, back when
it was Elena doing it, but the it seems we all forgot... Sorry for
that! :-/

I seriously think we should do something about this as, while in a non
vNUMA setup it can certainly cause weird/inconsistent performance, in a
vNUMA one, as shown, it's quite a hige mess.

- that's what our kernels have been doing for years, and
it seems like someone over here is now looking into whether this
could be done in pv-ops too (without too much uglification).

That would be great, IMO. I'd be up for helping with this, but I know
next to nothing about CPUID, so that would require some setup time. If,
at least, you could keep me in the loop it would be great.

In the meanwhile, what should we do? Document this? How? "don't use
vNUMA with PV guest in SMT enabled systems" seems a bit harsh... Is
there a workaround we can put in place/suggest?

I haven't been able to reproduce this on my Intel box because I think I have different core enumeration. Can you try adding
to your config file?

On AMD, BTW, we fail a different test so some other bits probably need to be tweaked. You may fail it too (the LLC sanity check).


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.