Improve Performance of FieldInterceptionHandler (and thus of flushing) ---------------------------------------------------------------------- Key: HHH-3909 URL: http://opensource.atlassian.com/projects/hibernate/browse/HHH-3909 Project: Hibernate Core Issue Type: Improvement Components: core Affects Versions: 3.3.1 Reporter: Ovidio Mallo Attachments: patch.txt In the FieldInterceptionHelper class, the method Class#getInterfaces() is used in conjunction with a subsequent loop over all interfaces of an entity class in order to check whether a bytecode instrumentation interface of CGlib or Javassist is found. This is done that way since CGlib and Javassist are both optional so they should only be accessed at runtime if available on the classpath. Attached you can find a simple patch which only once checks whether CGlib and/or Javassist is available and then uses the more performant "instanceof" operator in order to check whether an entity has an interceptor. The new code still guarantees that no CGlib or Javassist class is accessed if it is not on the classpath. The patch helps making the FieldInterceptionHelper methods more performant which can e.g. be noticed during a session flush, especially if bytecode instrumentation is used. In order to measure the performance impact of the patch I've written a simple program which does the following: 1. insert 100 thousand entities into a simple table with 6 numeric attributes 2. do an initial flush on the session in order to execute all the insert statements 3. loop 300 times and do a flush on the session (only a dirty check is performed, no database operation) I've time the flushes under point 3 above with and without bytecode instrumentation for the used entity. Here are the results: USING BYTECODE INSTRUMENTATION ============================= * SUN JVM 1.6.0_05: - without patch: 25734ms - with patch: 11015ms * IBM JVM 1.4.2: - without patch: 20672ms - with patch: 11140ms WITHOUT BYTECODE INSTRUMENTATION ================================ * SUN JVM 1.6.0_05: - without patch: 73328ms - with patch: 62015ms * IBM JVM 1.4.2: - without patch: 70344ms - with patch: 61125ms Using bytecode instrumentation, the performance improvement is about of a factor of 2-2.5. When not using bytecode instrumentation, the overall impact is of course less relevant since the rest of the work done during flush is more expensive anyway. In any case, IMHO the impact when using bytecode instrumentation is relevant. The patch is very simple and only touches the FieldInterceptionHelper class. -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://opensource.atlassian.com/projects/hibernate/secure/Administrators.... - For more information on JIRA, see: http://www.atlassian.com/software/jira