Performance and latency sensitive applications can be impacted by the default scheduling mechanism found in Openstack. the default OpenStack scheduler is optimized for system density: it attempts to launch a maximal number of VMs per host rather than optimizing the individual VM’s performance.
By configuring OpenStack to be NUMA aware VMs can see improved performance as their assigned CPUs have guaranteed memory access .
To support latency sensitive and high performance computing workload PF9 IaaS should support NUMA
https://www.stratoscale.com/blog/openstack/cpu-pinning-and-numa-awareness/