Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-19224

LNet should confirm TCP MTU between peers

    XMLWordPrintable

Details

    • Improvement
    • Resolution: Unresolved
    • Medium
    • None
    • Lustre 2.14.0, Lustre 2.17.0, Lustre 2.16.1
    • 3
    • 9223372036854775807

    Description

      There have been several instances of hard-to-debug issues in the network where clients have mismatched TCP/RoCE MTU.  This still allows "lctl ping" to work, if the ping packet is smaller than the MTU of any peer, but larger messages can fail to transmit and it causes intermittent issues in the system.

      It would be good for LNet to automatically detect and report an error if there is an MTU mismatch between peers in the same LNet network.

      Attachments

        Activity

          People

            wc-triage WC Triage
            adilger Andreas Dilger
            Votes:
            0 Vote for this issue
            Watchers:
            6 Start watching this issue

            Dates

              Created:
              Updated: