Re: [PATCH 2/4] libxc: report how much memory a domain has on each NUMA node

[George Dunlap <dunlapg@umich.edu>]

On Tue, Mar 11, 2014 at 7:04 PM, Dario Faggioli
<dario.faggioli@citrix.com> wrote:
> On mar, 2014-03-11 at 18:16 +0000, Ian Jackson wrote:
>> Dario Faggioli writes ("Re: [Xen-devel] [PATCH 2/4] libxc: report how much memory a domain has on each NUMA node"):
>> > In the current NUMA placement implementation, in libxl, what we need to
>> > know is how much memory is free on each node, rather than how much
>> > memory each domain occupies there.
>> >
>> > So, what is it that I should try to show you?
>>
>> I want a sketch of a race-free memory assignment approach that works
>> with ballooning (and domain self-directed numa migration), and doesn't
>> depend on arbitrating memory usage by having Xen fail certain memory
>> allocations.
>>
> Mmm... I think I see what you mean now (except perhaps for the "domain
> self-directed numa migration" part).
>
> I also still don't think that this specific call/functionality should be
> subject to (that much) locking, at least not at this level, but I see
> the value of having a plan.

I think it's worth taking a little bit of time seeing if we can make
an interface that could be used by a toolstack for such a purpose.
But if it's overly complicated, I don't see a problem with having one
interface just for information, and come up with a new race-free
interface once someone shows up that actually needs it.  Otherwise we
risk spending a lot of time inventing an API that turns out not to be
suitable for purpose anyway.

 -George