CPU Affinity on UFM-SDN Appliance

To execute major processes on the system (UFM, SM, SHARP, Telemetry), sub and multi-threaded processes are run on the system.

In order to increase UFM's efficiency, the number of context-switches is reduced. When each major CPU is isolated, users can decrease the number of context-switches, and the performance is optimized.

The CPU affinity of these major processes is configured in the following two levels:

  • Level 1- The major processes initiation.

  • Level 2- Preceding initiation of the model's main subprocesses.

According to user configuration, each process is assigned with affinity.

For using this feature,

  1. Run:

    Copy
    Copied!
                

    [CPUAffinity] Is_cpu_affinity_enabled=true Model_main_cpu_affinity=1-4 Sm_cpu_affinity=5-13 SHARP_cpu_affinity=14-22 Telemetry_cpu_affinity=22-23

  2. Check how many CPUs you have on your machine.

  3. Enable the feature on gv.cfg.

  4. Using the comments within the CPUAffinity section on gv.cfg define for each app the CPU set

  5. Restart UFM.

© Copyright 2023, NVIDIA. Last updated on Feb 8, 2024.