[Openstack] Anti-Affinity Based on NUMA Node

contact at ladenis.fr contact at ladenis.fr
Thu Jul 28 08:09:51 UTC 2016


On 27/07/2016 18:41, Chris Friesen wrote:
> On 07/26/2016 05:33 PM, Brent Troge wrote:
>> 
>> Instead of anti-affinity based on compute host, is there a way to 
>> filter based
>> on NUMA node ?
>> 
>> In my flavor I am using
>> 
>> hw_numa_nodes=1
>> 
>> However, for the VMs within the same server group, I would like to 
>> ensure they
>> do not end up in the same host NUMA node.
> 
> As far as I know vanilla OpenStack has no way to do this.
> 
> Chris
> 
> _______________________________________________
> Mailing list: 
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
> Post to     : openstack at lists.openstack.org
> Unsubscribe : 
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack

Hello,

you could consider a more common way of doing that. Use the 
NUMATopologyFilter scheduler and set the extra specs of your flavor or 
image that you want to be NUMA aware like so:

* hw:numa_nodes=NN - numa of NUMA nodes to expose to the guest.
* hw:numa_mempolicy=preferred|strict - memory allocation policy

You should define a flavor which fits on your NUMA node, like 12 VCPU 
and half of the total RAM for a 24 PCPU hypervisor. Like that every VM 
will have its dedicated NUMA node with local memory.
For HPC, we are doing so to be sure that the resources are not wasted 
and reduce the overhead of virtualization.

Regards,




More information about the Openstack mailing list