Re: [Xen-devel] [PATCH 0/4] mitigate the per-pCPU blocking list may be too long

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 26.04.17 at 05:30, <chao.gao@xxxxxxxxx> wrote:
> On Wed, Apr 26, 2017 at 02:19:22AM -0600, Jan Beulich wrote:
>>>>> On 26.04.17 at 02:52, <chao.gao@xxxxxxxxx> wrote:
>>> Patch 2/4 randomly distritbutes entries (vCPUs) among all oneline
>>> pCPUs, which can theoretically decrease the maximum of #entry
>>> in the list by N times. N is #pCPU.
>>
>>Why randomly? Shouldn't current list length determine which CPU(s)
>>to prefer?
> 
> I thought randomly distribution is simple and also can meet our
> demand.  What you said I think is also ok. How about just choose the
> CPU which has the shortest list length?

If that's doable without having to iterate over all CPUs every time.
There may need to be a hybrid approach, or one only coming close
to optimal distribution.

> I will use this policy in next
> version. Will you review this version when you have time?

Of course, but I can't predict when that will be.

> or I can
> prepare and send next version directly?

That's entirely up to you.

> Regarding to the test I do, any other complementary tests do you
> think are needed to prove that these patches really mitigate the
> issue?

I think the primary criteria is that you can abstractly prove the
problem to be gone. That's (as I'm sure you understand) an
argument against using any randomness as input here.

Jan


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