To enable vdpa with multi queue tehre are many aspect that need to be assets.

(1) Report the device queue to the db so that it can later be used for scheduling

(2) Ensure that the libvirt driver can properly configure multi-queue. the generic code might work but it may need to be adapted to clamp the number of core to min(flavor.cpus, device. queues)

(3) Request the number of queues for each device ideally as metadata on the port passed from neutron this may or may not require a neutron change but it could be a new QOS type or similar extension we would then need to modify the scheduler/pci claim code and or placement modeling to accommodate the new resource.

(4) Work to track VF and VDPA devices in placement is being done by the community this cycle. while we could do the queue aware scheduling on the nova side eventually it should be modeled in placement somehow, so there would be follow on work to enable that