This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
Home Products Support Community News


RE: [Xen-devel] Xen CPU limit?

> -----Original Message-----
> From: Pasi Kärkkäinen [mailto:pasik@xxxxxx]
> Sent: Tuesday, March 30, 2010 1:53 AM
> To: Jan Beulich
> Cc: xen-devel@xxxxxxxxxxxxxxxxxxx; Martin Lukasik
> Subject: Re: [Xen-devel] Xen CPU limit?
> On Tue, Mar 30, 2010 at 08:16:26AM +0100, Jan Beulich wrote:
> > >>> Pasi Kärkkäinen<pasik@xxxxxx> 29.03.10 17:49 >>>
> > >I think the number of virtual cpus (vcpus) per guest is (was) 32,
> > >and now in Xen 4.0.0 you can have up to 64 vcpus per guest.
> >
> > I don't think the tools are up to anything beyond 32 yet.
> Hmm.. what's actually missing from the tools to handle >32 vcpus?
> > The (64-bit) hypervisor allows up to 8192 vCPU-s iirc (but that's a
> truly
> > theoretical limit, as Dom0 or a guest likely won't be able to bring
> up
> > that many due to there only being 4096 event channels; current
> > Linux requires 5-6 of them per vCPU for IPIs and timer vIRQ).
> >
> Wow.. I wasn't aware the (theoretical) limit is that big :)

But, if I am reading this correctly, 4096 event channels
is the next scalability barrier.  If, on a "big" machine, one
tries to run 64 (nearly always idle) guests each configured
with 16 vcpus (because they run a busy database load when
they are not idle), it won't work due to the event channel limit?

This scenario seems quite possible in a cloud/hosting environment.

Xen-devel mailing list