Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-5777

Significant performance regression using TokuDB tables in mariadb.org builds

        Details

        • Type: Bug
        • Status: Closed
        • Priority: Major
        • Resolution: Fixed
        • Affects Version/s: 5.5.36, 10.0.8
        • Fix Version/s: 5.5.37, 10.0.10
        • Component/s: None
        • Environment:
          Ubuntu 13.10, (2) x Intel Xeon E5540, 48GB, 8 x 10K SAS in RAID10, 256MB Cache, XFS

          Description

          In testing the performance of TokuDB tables using the MariaDB builds (available from mariadb.org) I noticed significant performance regressions. The following is for Sysbench (data > cache, IO bound):
          workload = read/write
          tables = 16
          distribution = uniform
          rows / table = 10 million
          tokudb_cache_size = 8G
          tokudb_directio = 1
          tokudb_read_block_size = 64K
          tokudb_row_format = tokudb_lzma
          performance_schema=off

          Table are bulk-loaded, then ran the test for 64 and 128 client threads, 5 minutes per thread count. Results as follows:

          MariaDB 10.0.8 : source = mariadb.org
          64 threads, 78.0 tps
          128 threads, 95.6 tps

          MariaDB 5.5.36 : source = mariadb.org
          64 threads, 78.4 tps
          128 threads, 95.8 tps

          MariaDB 5.5.30 : source = tokutek.com
          64 threads, 132.3 tps
          128 threads, 161.5 tps

          MySQL 5.5.30 : source = tokutek.com
          64 threads, 134.8 tps
          128 threads, 164.1 tps

          MySQL 5.5.36 : source = tokutek.com
          64 threads, 131.2 tps
          128 threads, 161.4 tps

            Gliffy Diagrams

              Attachments

              1. global-variables.diff.txt
                5 kB
              2. oltp_ro_5.5.30.gif
                oltp_ro_5.5.30.gif
                10 kB
              3. oltp_ro_5.5.35.gif
                oltp_ro_5.5.35.gif
                10 kB
              4. series53.ods
                28 kB
              5. series54.ods
                23 kB
              6. tokudb_oltp.gif
                tokudb_oltp.gif
                14 kB

                Activity

                Ascending order - Click to sort in descending order
                10 older comments
                Hide
                Permalink
                tmcallaghan Tim Callaghan added a comment - - edited

                It appears that the COMPRESSION table create option does not pickup the tokudb_row_format default that is set for the session/global variable.

                Therefore, if the user does not use "COMPRESSION=" in their CREATE TABLE statements they are getting uncompressed, which is the current MariaDB default.

                Show
                tmcallaghan Tim Callaghan added a comment - - edited It appears that the COMPRESSION table create option does not pickup the tokudb_row_format default that is set for the session/global variable. Therefore, if the user does not use "COMPRESSION=" in their CREATE TABLE statements they are getting uncompressed, which is the current MariaDB default.
                Hide
                Permalink
                axel Axel Schwenke added a comment -

                Hi Joel, Tim!

                good catch! How embarrassing that I didn't notice the size difference for the data directories. Note to self: whenever looking at a performance problem with a compressing engine, check table space size first.

                Anyway: Serg has already fixed this in his working tree and it should soon be pushed to the main trees. For 5.5 the compression default will be zlib, for 10.0 the compression default will be taken from a session variable (which in turn is initialized from the global variable)

                Show
                axel Axel Schwenke added a comment - Hi Joel, Tim! good catch! How embarrassing that I didn't notice the size difference for the data directories. Note to self: whenever looking at a performance problem with a compressing engine, check table space size first. Anyway: Serg has already fixed this in his working tree and it should soon be pushed to the main trees. For 5.5 the compression default will be zlib, for 10.0 the compression default will be taken from a session variable (which in turn is initialized from the global variable)
                Hide
                Permalink
                tmcallaghan Tim Callaghan added a comment -

                I modified the KB page at https://mariadb.com/kb/en/tokudb-differences/, noting this behavioral difference.

                Show
                tmcallaghan Tim Callaghan added a comment - I modified the KB page at https://mariadb.com/kb/en/tokudb-differences/ , noting this behavioral difference.
                Hide
                Permalink
                tmcallaghan Tim Callaghan added a comment -

                Thanks for the update Axel, I wish the default could come from the session variable in both but zlib is much better than uncompressed.

                Show
                tmcallaghan Tim Callaghan added a comment - Thanks for the update Axel, I wish the default could come from the session variable in both but zlib is much better than uncompressed.
                Hide
                Permalink
                axel Axel Schwenke added a comment -

                Serg, please close this issue when you see fit.

                Show
                axel Axel Schwenke added a comment - Serg, please close this issue when you see fit.

                  People

                  • Assignee:
                    serg Sergei Golubchik
                    Reporter:
                    tmcallaghan Tim Callaghan
                  • Votes:
                    3 Vote for this issue
                    Watchers:
                    14 Start watching this issue

                    Dates

                    • Created:
                      Updated:
                      Resolved: