On Sun, Jun 7, 2009 at 1:42 AM, Michael Widenius<monty@askmonty.org> wrote:
Hi!
"MARK" == MARK CALLAGHAN <mdcallag@gmail.com> writes:
MARK> On Tue, May 19, 2009 at 1:26 PM, Giuseppe Maxia <g.maxia@gmail.com> wrote:
Hi, I am doing some tests with Maria. Out of the box, without any tuning, the load of records (with multiple record INSERT statements) is 5 times slower than MyISAM, 4 times slower than InnoDB.
MARK> I tested Maria with sysbench and it was many times slower than other engines.
Yes, if you are using transactions=1 and short records or a lot of updates/deletes it's slower than other engines because we are still using table locking, we don't have group commit and we need a better implementation of the key cache (less locking).
When running sql-bench (single threaded) we are faster than InnoDB on many of the tests, so at least some things are done right.
As you probably know, Maria 1.5 is about correctness, not speed. I belive that it's critical to first get things to work correctly before you starting to optimize.
We are working on fixing many of the issues in Maria 2.0; Sanja pushed group commit some time ago in MariaDB 5.2 and this should improve stuff a bit, but we know there is a long way to go before we are in par with InnoDB. We will get a better picture as soon as we have the transactional part in as then we can get rid of the table locking problem.
Regards, Monty
Are there any useful (to maria developers) performance tests we can run? If we limit all tests to 1 user (no concurrency) are the results useful? -- Mark Callaghan mdcallag@gmail.com