NUMA vertical alignment ensures that all CPU cores that a particular data path uses for network traffic are located on the same NUMA node. When a VNF-C uses the sched.cpu.latencySensitivity=HIGH parameter and is placed on a NUMA node, N-VDS Enhanced ensures that the VNF-C uses the logical cores from the same NUMA node. This function provides the fastest access between CPUs, their caches, and memory pages for optimal performance. This function also helps avoiding the inefficiencies of crossing the QuickPath Interconnect (QPI) bus between the NUMA nodes.

By setting sched.cpu.latencySensitivity to HIGH, N-VDS Enhanced ensures that the virtual machine, and logical cores are vertically aligned as long as resources are available. The VM-specific configuration is discussed in Accelerating Data Plane Workloads.

Recommendations

  • Replicate the NUMA node configuration in terms of NIC placement and logical cores assignment to N-VDS Enhanced across all NUMA nodes.
  • Set the latency sensitivity VM configuration to high for data plane intensive VNF-Cs.
  • NUMA node affinity setting is not required.