Looks like someone is flaming me for my criticism of OJB.  Just to set the record straight,  Dave Johnson's blog clarified the OJB JDO issue. I had this to say about JDO,  and this to say about OJB in comparison with Hibernate.

All of this is pretty old news, Matthew Porter and I were both users of OJB and we were both complaining about its state and direction last December.  Well, Matthew has moved on to Hibernate, so has Dave Johnson.  Meanwhile, I haven't had the chance to do any new OR mapping lately, but I'll possibly do that in Hibernate too. 

Again, I appreciate the need for "architectural beauty", patterns, extensibility, etc. that OJB is trying to achieve, unfortunately the pragmatic approach of  Hibernate will ultimately prevail in this situation.  That's because the OR mapping problem is not well understood, and devising an elegant solution is at best premature. Attempting to build a universal layer (i.e. OTM) to support ODMG and JDO, two otherwise flawed standards, is an exercise of futility.  Case in point, Hibernate today is almost completing its version 2.0, while OJB has yet to complete version 1.0.  

OJB isn't really that bad, in fact, I haven't replaced it in one of my projects. But it's not the best OR mapping tool on the planet, TopLink (which I've also worked with) is much better and Hibernate (just from observations) looks better.  In fact, Hibernate may in fact be even better than TopLink, but I'll know that when I actually use it. 

Now, you may think that OJB's architecture smokes Hibernate, well that's your "expert" opinion.  I think that this "architectural beauty" argument is irrelevant and more of a weakness, and that's my "expert" opinion.