Analyzing statement execution

After you create indexes, make sure that Derby is using them. In addition, you might also want to find out the join order Derby is choosing.

A general plan of attack for analyzing your application's SQL statements:

  1. Collect your application's most frequently used SQL statements and transactions into a single test.
  2. Create a benchmark test suite against which to run the sample queries. The first thing the test suite should do is checkpoint data (force Derby to flush data to disk). You can do that with the following JDBC code:
    CallableStatement cs = conn.prepareCall
  3. Use performance timings to identify poorly performing queries. Try to distinguish between cached and uncached data. Focus on measuring operations on uncached data (data not already in memory). For example, the first time you run a query, Derby returns uncached data. If you run the same query immediately afterward, Derby is probably returning cached data. The performance of these two otherwise identical statements varies significantly and skews results.
  4. Use RunTimeStatistics to identify tables that are scanned excessively. Check that the appropriate indexes are being used to satisfy the query and that Derby is choosing the best join order. You can also set derby.language.logQueryPlan to true to check whether indexes are being used or not. This property will is print query plans in the derby.log file. See derby.language.logQueryPlan, as well as Working with RunTimeStatistics for more information.
  5. Make any necessary changes and then repeat.
  6. If changing data access does not create significant improvements, consider other database design changes, such as denormalizing data to reduce the number of joins required. Then review the tips in Application and database design issues.
Related concepts
Application and database design issues
Working with RunTimeStatistics