Optimizing VM Placement

Running a VM on the same NUMA node as its NIC minimizes cross-node memory access and resource contention, significantly reducing network I/O latency while improving throughput.

In the NUMA architecture, local memory access is substantially faster than remote memory access. When vCPUs and memory of a VM are allocated on the same NUMA node as its NIC, data transfers (including network packet processing) occur entirely through local memory, eliminating the additional latency and bandwidth loss of cross-node communication.

1

cat /sys/class/net/<NIC_name>/device/numa_node

For example, querying the NUMA node for NIC enp65s0f0np0 reveals that it resides on node 0. Consequently, the vCPUs of the VM should be bound to physical CPU cores within NUMA node 0.