package-size | Get the bundle size of an npm package | Plugin library

my impression is, that hibernate.default_batch_fetch_size should work similar to setting the batch size in a jdbc-PreparedStatement. My experiments suggests this is not the case. How can I force Hibernate to use a larger fetch-size?

We have a spring/hibernate webapp using an Oracle RDBMS for persistence. We recently came across a situation, where we had a one-to-many association with more than one million entries on the "many" side.

It took Hibernate more than a minute to materialize the collection. Evaluation with a profiler shows, that hibernate executes one load statement and the time is lost during transfer of the data (not preparation of the data in the RDBMS). It is sort of suspicious, that the profiler reports a lot of socket reads which results in very small data packages. Also, every read-package has a corresponding write-package.

We tried to optimize hibernate.default_batch_fetch_size which was initially set to 4. However, increasing to 10 or even 100 made execution slower (thus proving, that the changes had some effect). Reduzing to 1 also slowed down the fetch. Socket-package-sizes were largely unaffected when changind the fetch size in Hibernate.

Bepuzzled I tried a plain Java program, which mimics what hibernate should be doing (i.e. setting the batch-size for the statement, constructing the resulting objects via reflection). Depending on the batch size I got execution times of less than 1 second.

Profiler supports the obvious conclusion: The read-packages grow more than tenfold in size and the number of write-packages is massivly reduced. The overall amount of data transfered is greatly reduced: Hibernate reads roughly 56MB from the socket, plain JDBC manages with large fetch size with less than 10MB. Reducing the fetch size on the JDBC side equalizes the picture.

Question: What do I have to do for Hibernate to make use of the JDBC-Fetch size? Obviously, simply setting hibernate.default_batch_fetch_size is not sufficient.

Edit, conclusion

With the suggestion of @TRW it was possible to speed up the load. Hibernate is still a lot slower than plain JDBC but we got a speedup of factor 3 and are still trying to find out the optimum setting of the parameters.