Reducing calls to __wt_epoch from session reset

XMLWordPrintableJSON

    • Type: Improvement
    • Resolution: Fixed
    • Priority: Major - P3
    • WT3.2.0, 4.1.9
    • Affects Version/s: None
    • Component/s: None
    • None
    • Storage Engines 2019-02-25, Storage Engines 2019-03-11
    • 5

      As analysing the performance of ycsb 100update workload on high performance machines, clock_gettime can degrade the performance in virtualization layer. This call in WiredTiger is called from __wt_epoch (which essentially comes from __wt_session_cursor_cache_sweep call stack) and needs to be investigated.

        1. __wt_epoch_overhead_portion_ycsb_100update-on-i3.metal.png
          __wt_epoch_overhead_portion_ycsb_100update-on-i3.metal.png
          52 kB
        2. __wt_epoch_overhead_portion_ycsb_100update-on-M200.png
          __wt_epoch_overhead_portion_ycsb_100update-on-M200.png
          81 kB
        3. __wt_epoch_portion_ycsb_100update-on-M200.after.png
          __wt_epoch_portion_ycsb_100update-on-M200.after.png
          80 kB
        4. __wt_epoch_portion_ycsb_100update-on-M200.before.png
          __wt_epoch_portion_ycsb_100update-on-M200.before.png
          108 kB
        5. diagnostic.data.reduce-epoch-reset.tar.gz
          879 kB
        6. diagnostic.reduce-cursor-sweep-reset.png
          diagnostic.reduce-cursor-sweep-reset.png
          375 kB
        7. ftdc-difference-cursor_sweep.png
          ftdc-difference-cursor_sweep.png
          234 kB

              Assignee:
              Hyeongwon Jang (Inactive)
              Reporter:
              Hyeongwon Jang (Inactive)
              Votes:
              0 Vote for this issue
              Watchers:
              9 Start watching this issue

                Created:
                Updated:
                Resolved: