Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-9749

Reduce overhead for ll_do_fast_read

    XMLWordPrintable

Details

    • Bug
    • Resolution: Fixed
    • Minor
    • Lustre 2.11.0
    • None
    • None
    • 3
    • 9223372036854775807

    Description

      For small read sizes (1k and below), creating a cl_env can be over 90% of the time spent doing ll_do_fast_read.

      ll_do_fast_read doesn't really need a cl_env, it only uses it for some debug printing. Not prepping a cl_env for ll_do_fast_reads improves read rates (for already cached data) for 1 byte reads by ~20x, and 1k reads by ~10x. 4k reads are improved by ~5x.

      Patch coming shortly.

      Attachments

        Activity

          People

            paf Patrick Farrell
            paf Patrick Farrell
            Votes:
            0 Vote for this issue
            Watchers:
            4 Start watching this issue

            Dates

              Created:
              Updated:
              Resolved: