By Mat Keep on May 29, 2012
Following the release of MySQL Cluster 7.2, the Engineering has been busy publishing a range of new performance benchmarks, most recently delivering 1.2 Billion UPDATE operations per Minute across a cluster of 30 x commodity Intel Xeon E5-based servers.
Figure 1: Linear Scaling of Write Operations
These performance tests have been run on the flexAsynch benchmark, so in the this blog, I wanted to provide a little more detail on that benchmark, and provide guidance on how you can use it in your own performance evaluations.
FlexAsynch is an open source, highly adaptable test suite that can be downloaded as part of the MySQL Cluster source tarball under the <storage/ndb/test/ndbapi> directory.
An automated tool is available to run the benchmark, with full instructions documented in the README file packaged in the dbt2-0.37.50 tarball. The tarball also includes the scripts to run the benchmark, and are further described on the MySQL benchmark page.
The benchmark reads or updates an entire row from the database as part of its test operation. All UPDATE operations are fully transactional. As part of these tests, each row in this benchmark is 100 bytes total, comprising 25 columns, each 4 bytes in size, though the size and number of columns are fully configurable.
Database access from the application tier is via the C++ NDB API, one of the NoSQL interfaces implemented by MySQL Cluster that bypasses the SQL layer to communicate directly with the data nodes. Other NoSQL interfaces include the Memcached API, Java, JPA and HTTP/REST
flexAsynch makes it possible to generate a variety of loads for MySQL Cluster benchmarking, enabling users to simulate their own environment.
Using the scripts in the dbt2-0.37.50 tarball and flexAsynch it is possible to configure a range of parameters, including:
- The number of benchmark drivers and the number of threads per benchmark driver;
- The number of simultaneous transactions executed per thread;
- The size and number of records;
- The type of operation performed (INSERT, UPDATE, DELETE, SELECT);
- The size of the database.
By using flexAsync, users can test the limits of MySQL Cluster performance. The size of the database used for testing is configurable, but given that flexAsynch can load more than 1GB of data per second into the MySQL Cluster database, users should ensure the database is large in order to achieve consistent benchmark numbers.
We will shortly publish a whitepaper that discusses MySQL Cluster benchmarking in more detail – so stay tuned for that. In the meantime, flexAsynch is fully available today for you to run your own tests. Also check out our Best Practices guide for optimizing the performance of MySQL Cluster