Re: [Xen-devel] [PATCH 1/5] x86: allow specifying the NUMA nodes Dom0 should run on

[Dario Faggioli <raistlin.df@xxxxxxxxx>]

On Fri, 2015-03-06 at 10:46 +0000, Dario Faggioli wrote:
> On Fri, 2015-03-06 at 09:11 +0000, Jan Beulich wrote:
> > >>> On 03.03.15 at 11:51, <JBeulich@xxxxxxxx> wrote:

> > Having implemented this "relaxed" addition (patch to be posted after
> > a few more tests), I find that with Dom0 being restricted to half of
> > the nodes of the test system, soft affinity set to that set, and hard
> > affinity left set to "all", many Dom0 vCPU-s nevertheless run on the
> > CPUs not in its soft affinity (and there's no other load on the system).
>
Oh, BTW, in my previous email I was asking about system load because,
while replying, I missed this: "there's no other load on the system"...
sorry! :-/

> > Is there a bug in that (credit) scheduler logic somewhere?
> > 
> There may be, of course, but nothing showed up during testing and
> benchmarking the feature. It's true that I probably concentrate mostly
> on DomU (especially while benchmarking), but it worked for me, and
> numbers from benchmarks confirmed that.
>
I'm testing soft affinity for Dom0 _without_ your patches, i.e., I'm
just setting soft affinity for Dom0's vCPUs after boot (hard affinity
set to "all") and looking at where they executes, both with and without
other load, and results look consistent to me.

I see Dom0 vCPUs executing almost only on pCPUs from their soft affinity
set, at least all the times that this is possible. If I generate other
vCPU load aimed at kicking them away from there, they do go away, but
they come back to such set as soon as the load disappears.

So, soft affinity per-se seems to be working for me.

I'll now apply your patch and see whether that changes thing (seems
unlikely, though).

If you want me to try replicate some specific testing scenario, feel
free to provide more details about it and I'll give it a go. :-)

Regards,
Dario

Attachment: signature.asc
Description: This is a digitally signed message part

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel