| |
|
 |
|
 |
|
|
|
| |
|
|
xen-users
RE: [Xen-users] Q: NUMA and XEN
> -----Original Message-----
> From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx
> [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of
> Ulrich Windl
> Sent: 09 October 2006 07:10
> To: xen-users@xxxxxxxxxxxxxxxxxxx
> Subject: [Xen-users] Q: NUMA and XEN
>
> Hi,
>
> I have a simple question, that's maybe an FAQ: The XEN
> hypervisor only reports one
> zonelist during boot, while a normal kernel will report two
> for a two-node NUMA
> machine. Now I'm wondering if I create two VMs with roughly
> half of the RAM
> (amount one node has), will XEN be clever or dumb when
> assigning RAM to those VMs?
Dumb. There's no guarantee that any domain will get any particular
memory, or even that the memory is contiguous within a lump. Until the
actual code to support NUMA is in place, you're just as well off letting
the two DomU's switch between CPU's as they like - that will at least
average out, rather than being "always bad" (if you're unlucky).
Of course, on a two-node Opteron system, I doubt that you'll be
measuring a huge difference unless you're running some HPC application -
in which case you should consider "not using Xen", as HPC on Xen isn't
the most optimal solution [lots of things are slowed down a little bit
by the extra layering - which isn't exactly great for your average HPC
application].
>
> XEN will be clever if it takes the RAM from one node,
> preferrably that where the
> assigned CPU is, and it will be dumb if it assigns just any
> RAM, causing a
> slowdown of RAM access.
Since Xen isn't aware of which RAM belongs to which processor, it will
be completely random whether you get "good" or "bad" result.
--
Mats
>
> Regards,
> Ulrich
>
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-users
>
>
>
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
|
|
| |
|
Home