Uploaded image for project: 'WiredTiger'
  1. WiredTiger
  2. WT-2439

Enhance reconciliation page layout

    • Storage 2017-03-27, Storage 2017-04-17

      Especially for workloads with updates that grow records. For example I ran a wtperf workload with the following configuration (requires changes for WT-2366):

      conn_config="cache_size=2GB,checkpoint=(wait=60)"
      table_config="type=file,leaf_page_max=32k,leaf_value_max=128k,split_pct=90"
      # The values are starting small, insert a lot so our database grows larger than
      # cache quickly.
      icount=20000000
      report_interval=5
      run_time=3000
      populate_threads=1
      # Continue inserting new records.
      threads=((count=1,inserts=1,throttle=1000),(count=4,update=1,update_delta=100))
      # Start with small values and let them grow slowly to large values.
      value_sz=20
      value_sz_max=65536
      

      After the test has run for half an hour the database ends up with an unbalanced distribution of record counts per leaf page (picture to follow).

        1. 2gb_80_85.png
          44 kB
          Sulabh Mahajan
        2. change_28_02_results.png
          80 kB
          Sulabh Mahajan
        3. page_histogram.png
          136 kB
          Alexander Gorrod
        4. poc_results.png
          69 kB
          Sulabh Mahajan
        5. workload_split_pc_75.png
          66 kB
          Sulabh Mahajan
        6. workload_split_pc_90.png
          76 kB
          Sulabh Mahajan
        7. wt_dump_load_90.png
          67 kB
          Sulabh Mahajan

            Assignee:
            sulabh.mahajan@mongodb.com Sulabh Mahajan
            Reporter:
            alexander.gorrod@mongodb.com Alexander Gorrod
            Votes:
            0 Vote for this issue
            Watchers:
            4 Start watching this issue

              Created:
              Updated:
              Resolved: