Percona-XtraDB-9.1: released and new coming features

PREVIOUS POST
NEXT POST

Recently Alexandr announced new Percona-XtraDB-9.1 release, and now it is good time to summarize features we have and what is expected soon.

This release contains long waited features from 5.0:

  • extended slow.log
  • USER/TABLE/INDEX/CLIENT_STATISTICS + THREAD_STATISTICS ( coming in release-10)

Extended slow.log is now even more extended, there is additional information for each query:

That is you can see how many bytes was returned by query, was temporary table used,
was it disk table or in-memory, and how big was temporary table.

Also you can profile each individual query from stored procedures, so now you should not
guess what took so long in stored procedure.

What we are working on right now is:

  • InnoDB pagesize, so you will be able to specify 4K, 8K, 16K pagesize when you created database
  • InnoDB fast checksums, idea and implementation was taken from Facebook patches
  • InnoDB big log files ( bigger 4GB). As our benchmarks on FusionIO shows, you may need
    4GB+ log files to get more performance on fast storage systems
  • SHOW TEMPORARY TABLES, patch from Venu Anuganti
  • INFORMATION_SCHEMA.SYS_TABLES and SYS_INDEXES to show internal InnoDB data dictionary

Full list of XtraDB engine features (what makes it different from InnoDB-plugin) on the state as it is right now:

Performance improvements

  • Improved buffer_pool scalability
  • Fast recovery
  • Improved IO path
  • Improved rollback segment scalability
  • Separate purge thread
  • Limited size of data dictionary in memory
  • Increased number of concurrent write transactions (undo slots) ( up to 4000 )
  • Fast checksums ( in release process )
  • Support of different pagesizes ( 4K, 8K, 16K) ( in release process )

Usability / operations

  • Show content of buffer_pool
  • Import / export of dedicated tables
  • Import / export of buffer_pool
  • Transactional replication
  • Show internal InnoDB data dictionary
  • Show InnoDB locking/io profiling in slow.log

I hope you enjoy our work!

PREVIOUS POST
NEXT POST

Comments

  1. says

    Yes. No changes are needed — it Just Works, and so will any future additions, as long as they follow the simple and predictable syntax of the slow log.

  2. Robert Wultsch says

    Has Percona considered creating a server variable (either static or dynamic) to disable automatic deadlock detection?

  3. says

    Robert,

    It is not in our plans, but sounds as good request.
    Do you mind to fill bug report ? I think we can implement it soon.

  4. says

    Than you for your work – we are using it now on magentocommerce.com and speed improvement is really great. Also, query profiling and debugging is much easer then with vanilla MySQL.

  5. says

    Hi,
    I’m testing this release on redhat5 64bits and so far the performance gain from a vanilla 5.0 is roughly 10x, I’m very impressed, thx a lot for the work done. One thing I can not get to work is the INDEX_STATISTICS table and the TABLE_STATISTICS one, the USER_STATISTICS is running fine after setting the userstat_running global variable ON but the other 2 tables are still empty. Is this working on this release?

    (Server version: 5.1.43-51-log Percona SQL Server (GPL), XtraDB 9.1, Revision 57)

    btw. we are planning to go in production with this release for bligoo.com (5.5 million visitors/month)

    king regards,
    Emilio

  6. Wilfried Schobeiri says

    Are you guys planning on updating your rhel builds to include 9.1? Seems the MySQL-percona builds are still on XtraDB 9.

  7. Timothy Denike says

    Any idea if the bugfix for 49047 will be merged into the 5.0 release? We’re running percona highperf 5.0.84 version currently and getting bitten by this bug occasionally during high concurrency updates to a single row. (UPDATE blah set foo=foo+1 where id=12345;)

Leave a Reply

Your email address will not be published. Required fields are marked *