Major Ensure that multiple instances of partner application (represented by Pods) are placed equally across compute die's on the same processor (See diagram below) so that one compute die doesn't get overloaded while others run idle.
The INTEL CPU at hands is called Clearwater Forest, single socket, in SNC-3 mode (Sub-NUMA Clustering 3).
Example:
A pod, pod#1, is scheduled to run.
The NUMA-aware scheduler select the appropriate worker node: Worker node 0 is selected.
Then the Kubernetes Topology Manager select the appropriate NUMA Zone so that all the required CPU comes from the same NUMA Zone: NUMA Zone 0 is selected.
A subsequent pod, pod#2 is scheduled.
The NUMA-aware scheduler select the appropriate worker node: Worker node 0 is selected.
Then the Kubernetes Topology Manager select the appropriate NUMA Zone so that all the required CPU comes from the same NUMA Zone: NUMA Zone 1 or Zone 2 should be selected ensuring resources are properly utilized when multiple pods are deployed on the same worker node. NUMA Zone 1 is selected.
Possible reference:
