Re: [Xen-devel] Supported #of CPUs/VMs per CPUs

["George Dunlap " <dunlapg@xxxxxxxxx>]

100 virtual machines seems on one box seems like a lot to me. :-)

I beleve the Potemkin Virtual Honeyfarm project
(http://www.cs.ucsd.edu/~savage/papers/Sosp05.pdf) did some mods to
allow the numbers to go up into the multiple thousands.

-George

On 9/22/06, Petersson, Mats <Mats.Petersson@xxxxxxx> wrote:
> > -----Original Message-----
> > From: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx
> > [mailto:xen-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of
> > Florian Kirstein
> > Sent: 22 September 2006 01:34
> > To: xen-devel@xxxxxxxxxxxxxxxxxxx
> > Subject: Re: [Xen-devel] Supported #of CPUs/VMs per CPUs
> >
> > Hi,
> >
> > > There is no maximum number of domains per cpu.  The maximum
> > number of
> > > domains *per machine* (regardless of the # of CPUs) depends on the
> > > amount of resources available, mainly memory.
> > > I've never tried to max out the numer if idle domains in Xen, but
> > > in general, the answer is quite a bit.
> > I did try it out (see "maximum Number of DomUs" posting a few months
> > back) on a 16 GB Quad-Opteron, and was surprised hitting a limit
> > way before I used up all my memory, at about 107 DomUs. Xen has a
> > private heap of 16MB and each Domain needs some memory there, so
> > (possibly depending of what else is stored there) at about 100 DomUs
> > that's it.
>
> And for x86_64 there is no reason you can't change this limit, ether by
> using the command-line argument, as I described in the post at that
> point, or changing the parameter in the relevant header-file (also
> described at the same time). There's really no reason why you can't
> change this if you need to increase beyond this number of domains.
>
> For 32-bit guests, it's a compromise, because the address space is
> cramped with "only" 4GB of space, because the special heap needs to be
> accessable by Xen at all times, thus mapped into all domains. This means
> that the larger it is, the more memory space is occupied by the heap,
> the less there is available for other things... I'm sure Keir or someone
> will be able to give more details of why it is a bad thing to make it
> much larger in 32-bit (including PAE, since PAE is only allowing the
> entire system to have more than 4GB of memory, but at any given time,
> the addressable space is 4GB, and the Xenheap must be mapped to ALL
> guests).
>
> In .../xen/include/asm-x86/config.h:
> XENHEAP_DEFAULT_MB    - Number of mbytes of heap. In 32-bit, this is 12
> from DIRECTMAP_MBYTES, and 16 for x86_64.
>
> Command line paramaeter:
> xenheap_megabytes
>
> --
> Mats
> >
> > Sure, not a real problem (there are few usefull szenarios in which
> > one would want 100 DomUs on one host), but worth mentioning IMHO...
> > Or did something change in this respect?
>
> It's been like that for some time, I don't know how long tho'. There is,
> as far as I understand, no STRICT limit for these things, it's just
> about compromises between one benefit and another.
>
> --
> Mats
> >
> > (:ul8er, r@y
> >
> > _______________________________________________
> > Xen-devel mailing list
> > Xen-devel@xxxxxxxxxxxxxxxxxxx
> > http://lists.xensource.com/xen-devel
> >
> >
> >
>
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-devel
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel