Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-23855

InnoDB log checkpointing causes regression for write-heavy OLTP

    XMLWordPrintable

Details

    Description

      Workload: sysbench OLTP write-only, i.e. as it is used by the regression benchmark suite in t_writes_innodb_multi.

      Setup: 16G buffer pool. 4G redo log. 4G or 8G data set. innodb_flush_neighbors = 0, innodb_io_capacity = 1000 (or 5000, 10000)

      Observation: after starting high, performance drops after ~ 1 minute. If waiting long enough, one can see oscillations in throughput. This seems to be related to Innodb_checkpoint_age reaching Innodb_checkpoint_max_age. There seems to be no LRU flushing at all, only flush_list flushing.

      Attachments

        1. Screen Shot 2020-10-20 at 5.59.20 PM.png
          Screen Shot 2020-10-20 at 5.59.20 PM.png
          690 kB
        2. oltp_ts_8.png
          oltp_ts_8.png
          26 kB
        3. oltp_ts_64.png
          oltp_ts_64.png
          41 kB
        4. oltp_ts_64.png
          oltp_ts_64.png
          27 kB
        5. oltp_ts_256.png
          oltp_ts_256.png
          39 kB
        6. oltp_ts_128.png
          oltp_ts_128.png
          30 kB
        7. Innodb-max-dirty-pages-pct-lwm.png
          Innodb-max-dirty-pages-pct-lwm.png
          60 kB
        8. Innodb_io_capacity.png
          Innodb_io_capacity.png
          66 kB

        Issue Links

          Activity

            People

              marko Marko Mäkelä
              axel Axel Schwenke
              Votes:
              0 Vote for this issue
              Watchers:
              14 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved:

                Git Integration

                  Error rendering 'com.xiplink.jira.git.jira_git_plugin:git-issue-webpanel'. Please contact your Jira administrators.