Uploaded image for project: 'WiredTiger'
  1. WiredTiger
  2. WT-10956

Investigate performance change in test/format after mirror branch (zseries)

    • Type: Icon: Task Task
    • Resolution: Unresolved
    • Priority: Icon: Minor - P4 Minor - P4
    • None
    • Affects Version/s: None
    • Component/s: None

      While investigating WT-10435, I noticed that the performance of test/format with mirroring not enabled regressed. This was done by comparing running with the config from WT-10435 (see the comments) with commit ffba1bcf0f15a28607c3b664d63d730149c1b3e and then with the mirror commit (addbbad09e823a5c1ed8407eb1e7b861e2bead6e).

      I only compared x86 and zseries and found interesting results (average from 5 runs):

      Pre mirror (x86): ~262 seconds

      Pre mirror (zseries): ~130 seconds

      Post mirror (x86): ~197 seconds

      Post mirror (zseries): ~168 seconds

      Yes it got faster on z86 but slower on zseries. More tests are needed but on zseries it was the difference to hitting WT-10435 or not (runs.timer=4 was set instead of runs.ops).

      Might want to check other platforms as well.

            Assignee:
            backlog-server-storage-engines [DO NOT USE] Backlog - Storage Engines Team
            Reporter:
            mick.graham@mongodb.com Mick Graham
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

              Created:
              Updated: