Verifying ptp4l Synchronization Accuracy

The following is an example of ptp4l output:

worddav798e39d2ba8c6604022447b529e0f366.png

Explanation of the columns:

  • rms: Root Mean Square of time offset from leader clock [nanoseconds]

  • max: Maximum time offset from leader clock [nanoseconds]

  • freq: Mean frequency offset from leader clock [ppb]

  • +/- (Freq): standard deviation of frequency offset from leader clock

  • delay: Mean path delay from the leader clock

  • +/- (delay): standard deviation of path delay from leader clock

PTP is considered stable if rms changes do not varying beyond 8 nsec, and if their absolute value is 2 nsec when there is no traffic.
To check if ptp4l is stable, the RMS should not pass 16 nsec.

Self-reported measurements from ptp4l are suspect to software noise and lack of precision. Instead, one can directly assess the PHC synchronization by using the ConnectX capability to raise an electrical signal on the PPS Out pin on a full second lapse. Using this method, the signal is raised using HW mechanisms, eliminating any inaccuracies. For details on how to enable PPS Out, please refer to Building and Installing ptp4l from Source.

© Copyright 2023, NVIDIA. Last updated on May 23, 2023.