Users are reporting CPU related performance problems on the virtual machine Email-Prod throughout the day. The resource settings for the VM and the ESXi host the VM is running on are shown below: VM:

Numa.PageMigEnable controls proactive page migration between nodes for memory balancing, but alone it does not address the scheduling imbalance causing the observed CPU performance problem.

Numa.AutoMemAffinity instructs the VMkernel to automatically migrate the VM's memory pages to the NUMA node where its vCPUs are currently scheduled, reducing costly cross-node memory access latency.

Setting CPU affinity to pin 2 vCPUs to each NUMA node ensures the VM's workload is explicitly balanced across both nodes, preventing all vCPUs from being co-scheduled on a single overloaded node while also enabling the memory scheduler to align memory with local vCPUs.

Binding all 4 vCPUs to a single NUMA node concentrates the entire workload on one node, squandering the second node's CPU resources and potentially worsening contention on that node rather than alleviating it.