SHARP Cleanup

NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP) Rev 3.4.0

This feature enables cleaning up all SHARP-related definitions when it is no longer desired to operate with it. The cleanup helps in leveraging the full potential of the switch capabilities without allocating resources for SHARP.

Furthermore, when an error takes place due to stuck jobs, uncleaned memory, or other scenarios, a cleanup should help in solving the error without the need for a switch reboot.

To perform a cleanup, follow the steps below.

  1. Stop sharp_am or make sure sharp_am is not running.

  2. Verify that there are no active SHARP jobs running. In case there are, be aware that cleaning SHARP resources will terminate these jobs, so either wait for them to finish or stop them gracefully.

  3. Run sharp_am with the following parameters: sharp_am --log_verbosity 3 --clean_and_exit TRUE

    sharp_am will clean all ANs mentioned in the smdb file and exit.

  4. To verify success, look into the sharp_am log file for a message in the following format: "Sent clean command to <> ANs, successes: <>, fails: <>"

© Copyright 2023, NVIDIA. Last updated on Nov 7, 2023.