Resource Management in NUMA Architectures

You can perform resource management with different types of NUMA architecture.

With the proliferation of highly multicore systems, NUMA architectures are becoming more popular as these architectures allow better performance scaling of memory intensive workloads. All modern Intel and AMD systems have NUMA support built into the processors. Additionally, there are traditional NUMA systems like the IBM Enterprise X-Architecture that extend Intel and AMD processors with NUMA behavior with specialized chipset support.

Typically, you can use BIOS settings to activate and deactivate NUMA behavior. For example, in AMD Opteron-based HP Proliant servers, NUMA can be deactivated by activating node interleaving in the BIOS. If NUMA is activated, the BIOS builds a system resource allocation table (SRAT) which ESXi uses to generate the NUMA information used in optimizations. For scheduling fairness, NUMA optimizations are not activated for systems with too few cores per NUMA node or too few cores overall. You can modify the numa.rebalancecorestotal and numa.rebalancecoresnode options to change this behavior.