Lower sizes for Oracle-specific row pre-fetch and statement cache.
Matthew have been advising me (around March 2004) that Oracle reports 10-15% improvements with their row pre-fetching. I have also heard from a DBA attending an Oracle seminar that Oracle recommend row pre-fetch with a value of 200.
However, it seems that OJB does not benfit too much from this and the memory requirements are just skyrocketing when the JDBC driver starts to prefetch for all it's worth. A customer of my day-job company were also getting frequent OutOfMemoryException in production with OJB 1.0.1 and Oracle row-prefetch=100, this disappeared when lowering this value to 10.
I know that Oracle row pre-fetch is complimentary to OJB proxy pre-fetch, so both should be kept in place. However it seems that OJB gets a much lower footprint without loosing noticeable performance if the values are lowered. Today I did some extensive testing and you can read the results here (no, I am not a graphics designed and I can't do good screenshots - click the method names if you dare...). :-)
http://www.apache.org/~mkalen/ojb/broker-tests.html
Values before commit are boxed in red, current values in green.
Unfortunately I was seeing the occasional failure in testRuntimeMetadataChanges (click the links from # failures), that we were previously seeing here:
http://mail-archives.apache.org/eyebrowse/ReadMsg?listId=107&msgNo=11589
After the latest Torque vs Oracle DATE/TIMESTAMP discussions I'm suspecting that this could in fact be related to a timing issue due to some DATE column in the Torque DDL, rather than any Oracle-specific optimizations in 9i platform. It is *not* reproducable to first run with PlatformOracleImpl and the PlatformOracle9iImpl (as I thought when I wrote the mail on the user-list), since it only fails in about 1 out of 10 runs even with 9i...
Cheers, Martin
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
