Which element affects the performance of virtual machines in NUMA architecture?

In Non-Uniform Memory Access (NUMA) architecture, the performance of virtual machines is significantly influenced by the affinity between virtual CPUs (VCPUs) and the physical nodes on which they are running. In a NUMA setup, each processor (node) has its own local memory, and accessing local memory is faster than accessing memory that's remote or located on another node.

When a VCPU is scheduled to run on a specific physical node and utilizes the memory that is local to that physical node, it benefits from reduced latency and increased throughput. This architecture allows VMs to perform optimally when their processing and memory needs are effectively managed in relation to the physical layout of the system.

In contrast, if VCPUs are frequently moved between nodes or access memory that is located on remote nodes, this can lead to performance bottlenecks, commonly known as "remote memory access penalties." Thus, ensuring a good affinity between VCPUs and their associated memory in the context of NUMA architecture is essential for maximizing the performance of virtual machines.