Performance comparison test for MySQL 5.6 and MariaDB-10.0

Performance comparison test for MySQL 5.6 and MariaDB-10.0time 2013-02-14 10:11:34 Open source China Original http://www.oschina.net/question/12_90065 ThemeMariaDBOLTPTesting Technology

Oracle has just released the MySQL 5.6.10 GA version, so it's time to update the previous performance test data, which includes the following versions:

• MySQL-5.5.29
• MySQL-5.6.10
• mariadb-5.5.28a
• MariaDB-10.0.1

The test also retains the 5.5 release for regression testing. We have often found that the new version lags behind in performance.

The test is performed in a different environment, with the main difference being the use of high-performance RAID-5 storage with 512 Gigabit battery-backed cache instead of SSDs. In addition, the tested machine has 16 cores, of which 12 cores run mysqld and the other 4 cores run Sysbench.

The test uses sysbench-0.5 OLTP, which packs 8 tables and 10G of data. The buffer pool size of InnoDB is 16G, log 4G. Different disk systems require different InnoDB configurations:

• Innodb_io_capacity = (is 20000 for SSD)
• Innodb_flush_neighbors = 1 (is 0 for SSD)

Here is the test result, first of all OLTP Read only:

It's very strange that MySQL 5.6 actually behaves abnormally in this round of testing. The difference between the 8 threads is small, and the 16 threads are best when they are different. However, the performance of higher concurrency drops quickly, even worse than MySQL 5.5. And the MariaDB 10.0 is worse than MariaDB 5.5, but not so obvious.

The response time chart represents a better and smoother:

MySQL-5.6 and MariaDB-10.0 look slightly better, meaning they can better allocate CPU cycles.

Disclaimer: This test does not use a thread pool. Oracle's thread pool implementation is already closed, so it's not possible to test and use it, which is a bit unfair if you use a thread pool on MariaDB.

If you want to understand the performance impact of the thread pool, you can view the previous two articles:

• MariaDB-5.5 Thread Pool Performance
• MariaDB-5.5 performance on Windows

Second Test: OLTP read-write test

Similar to a test in front of this figure, MySQL 5.6 and MariaDB 10.0 performance is much lower than the 5.5 version, in the case of high load, the decline of about 10%.

It is a well-known fact that MySQL 5.5 is under high load because of the performance slowdown caused by synchronous flush operations.

The response time graph is relatively better:

This is good news, and the 5.5 version responds a lot in the case of 64 threads or more threads. The Adaptive Flush algorithm for MySQL 5.6 and MariaDB 10.0 seems to work well.

Here's another question: if you're using multiple buffer pool instances, you'll see more delay in write operations. In the above results, the read-only test used 16 buffer pools, while the read-write test used only 1.

Conclusion:

• MySQL-5.6 performance is worse than the previous version, especially in cases of high concurrency. This does not match the test results published by Oracle. I can only speculate on why the results differ so much, I guess, is the Oracle closed-source thread pool and Oracle testing on larger machines.
• There is no need to worry about write latency when using a single buffer pool. MySQL 5.6 also allows up to 512G of redo logs to reduce synchronous flush operations.

The script for this test can be accessed from the following address:
Http://bazaar.launchpad.net/~ahel/maria/mariadb-benchmarks/revision/20

You are welcome to re-do this test and share the test results with us.

Performance comparison test for MySQL 5.6 and MariaDB-10.0