Uploaded image for project: 'Blazegraph (by SYSTAP)'
  1. Blazegraph (by SYSTAP)
  2. BLZG-641 Improve load performance
  3. BLZG-1667

Growth in RWStore.alloc() cumulative time (but latency looks ok)

    XMLWordPrintable

    Details

    • Type: Sub-task
    • Status: Done
    • Priority: Medium
    • Resolution: Done
    • Affects Version/s: None
    • Fix Version/s: BLAZEGRAPH_2_0_0
    • Component/s: None
    • Labels:
      None

      Description

      We have observed the RWStore.alloc() latency growing faster than the time to write on the write cache service. This suggests a non-linear component in how allocations are made. With respect to the images, Journal / write secs is the time inside of RWStore.alloc() (while the lock is held) while Journal / write cache / write seconds is the time to write on the channel. The 2nd chart shows the elapsed time to insert a record into the write cache service, which is basically flat. Thus the growth in RWStore.alloc() time is NOT related to disk latency.

        Attachments

        1. image.png
          image.png
          11 kB
        2. image (1).png
          image (1).png
          13 kB
        3. image (2).png
          image (2).png
          11 kB

          Activity

            People

            Assignee:
            martyncutcher martyncutcher
            Reporter:
            bryanthompson bryanthompson
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

              Dates

              Created:
              Updated:
              Resolved: