It is my understanding that, with numa=on, Xen will try to allocate
memory for a para VM from the same node as the CPUs that the VM's procs
are pinned to.
Thats true for HVM guests, too.
Thus, in a multi-node system, if we specify a pinning to
a specific node's CPUs in a VM's config file, Xen
will allocate the VM's initial memory from that same node at startup.
Correct.
My question is: does this location preference happen for all subsequent
memory allocation? In other words, if I boot a VM configured with some
initial memory and pinned to CPUs in node-X, change the pinning of the
VM's procs to CPUs in node-Y, and then use "xm mem-set" to allocate more
memory, does this new memory come from node-Y?
AFAIK ballooing and NUMA are in the moment mutually exclusive.
Currently I am working on NUMA PV guest support (multiple nodes within a
guest). I posted preliminary patches for HVM guests in August.
If you want to help, you are invited ;-)
Regards,
Andre.
--
Andre Przywara
AMD-Operating System Research Center (OSRC), Dresden, Germany
Tel: +49 351 277-84917
----to satisfy European Law for business letters:
AMD Saxony Limited Liability Company & Co. KG
Sitz (Geschäftsanschrift): Wilschdorfer Landstr. 101, 01109 Dresden,
Deutschland
Registergericht Dresden: HRA 4896
vertretungsberechtigter Komplementär: AMD Saxony LLC (Sitz Wilmington,
Delaware, USA)
Geschäftsführer der AMD Saxony LLC: Dr. Hans-R. Deppe, Thomas McCoy
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
Home