Hello,

I was tempted to post this to the "Performance-Challenge", but the problem is so general that I think I must be missing something.

I've been testing Hibernate using the TPC-W benchmark (http://www.tpc.org), specifically, against the JDBC-Servlets implementation on jmob (http://jmob.objectweb.org/tpcw.html).

I'm using hibernate 2.0.3 and mySQL 4.0.15 here, with the Connector/J driver (it could be that the problem is specific to this driver, but even so, the most common opensource O/R mapper should work with the most common driver for the most common opensource DB, shouldn't it?)

Using JProbe to analyse the performance, I see that about 30%
of the time is spent in ResultSet.findColumn(String), which is called from ResultSet.get*(String). The problem is, getting a named column from a ResultSet is much more expensive than getting a numbered column. That is, ResultSet.getString("title") is slower than ResultSet.getString(1). Worse, since this extra work is associated only with a particular result set, the same overhead is incurred everytime an object is loaded. When we load a bunch of objects using JDBC that give the number of the column, we will always be faster than hibernate, since hibernate uses the name of the returned field.

With this information, it's pretty easy to generate an example where
jdbc outperforms hibernate. If needed, I can send the code I am using. I'm not sure I can post it, as I'm not the holder of the copyright (everything is GPL'd).

Am I missing something here, or is this a performance problem with Hibernate?

Best Regards,

Ron

net.sf.hibernate.test.PerformanceTest does not show this to be a problem, even on MySQL.

If you can really demonstrate such a big difference in performance, I would certainly like to know about it. (I'm not sure it quite qualifies for the $100, 'cos I say "must be reproduceable on >1 platform".)

So, send me something, since this is a different result to my tests.....

Hi,

I'm not interested in the $100, I'm more interested in helping to improve Hibernate. I am writing an article for a German magazine that uses the TPC-W benchmark to measure the performance of various software architectures, and would like to verify Hibernate's claim of being "very high performance".

My code base is too large for posting, I can send you an archive of the code, complete with scripts to populate the database, but I need an e-mail address to which to send it.

Ron

send it to gavin@hibernate.org

TIA