Because packet processing uses CPU cycles, the host CPU resources play crucial role in the network performance. To provide predictable throughput performance for data plane intensive workloads, N-VDS Enhanced uses dedicated CPU cores that are assigned to the switch during its creation. These dedicated CPU cores are referred to as logical cores. When you assign logical cores to N-VDS Enhanced, these cores are removed from the available CPU resources on the host. As a result, there is a reduced number of CPU cores and therefore host-level CPU cycles that are available to DRS and other vSphere.

You assign logical cores to N-VDS Enhanced per NUMA node. You assign the same number of logical cores to each NUMA node. This best practice allows the optimal utilization of the cluster that is dedicated to data plane intensive workloads.
Figure 1. NUMA Node Design for Accelerated Workloads

NUMA Node Design for Accelerated Workloads

The number of CPU cores assigned to N-VDS Enhanced depends on the speed of the physical NICs and the workload type. For example, an N-VDS Enhanced using 10GbE NIC and serving typical Web-based services where the traffic consists of packets larger than 1,000 bytes, can completely saturate the NIC with only one single logical core assigned to N-VDS Enhanced. Because each environment is different and workloads have unique characteristics, the exact number of logical cores assigned to N-VDS Enhanced must be evaluated by the CSP.

Recommendations

  • Assign logical cores to N-VDS Enhanced based on the expected traffic volumes, the servers' NIC speeds and the VNF's traffic characteristics.
  • Assign the same number of logical cores to N-VDS Enhanced on each NUMA node.