Is there any way to have Hibernate cache an entire object, complete with its collections, in a single cache? Rather than having a cache for the base object, association, child object, etc. We use a distributed replicated cache and having child objects potentially partitioned across 6 servers may be a huge performance hit. Would like to retrieve the entire object graph from one cache.

Thanks in advance for any help.

This is not how a distributed replicated cache works.

My mistake, I meant to say we use a distributed cache. In this case the data IS partitioned across multiple machines.

Regardless, is there any way to cache only "composed" objects through the Hibernate API? On the surface there does not appear to.

Hm, it is not possible and I still don't see why you would need this.

For an example we might have a listing, with say 20 items, each item with 5 attributes. In the distributed cache each object potentially (in the average case) incurs a slight network overhead. So if I could store the entire object in one cache, I would never incur more than one network hop. Now, the way I am using Hibernate now, I would incur 1 (listing) + 20 (collection association) + 20 (items) + 100 (collection association) + 100 (attributes) = 241 network hops. Suddenly even a slight network latency is magnified!

Obviously for some types of objects I can lazy load them and greatly reduce the overhead. However some will HAVE to utilitze their entire collection. Those I would like to store as one blob of an object rather than their constituent parts.

Worst case scenario I have to manually cache these types of objects. However the Hibernate API is simple and flexible and I would like to use it across our whole domain model.

Thanks for your help in any case.

A distributed cache doesn't have to do distributed lookups, just distributed invalidation or replication.

A replicated cache (ie complete data set on each cluster member) does not have a lookup penalty. A distributed (ie data set partitioned across n servers) will have to go across the network in many cases to retrieve the date. Our data set is much too large to be completely contained on a single cluster member.

Irregardless it sounds like there is no way to configure Hibernate for the type of caching I'm looking for.[/i]

I think that this is outside of the scope of Hibernate: Hibernate will use a cache region for each class and each collection that should be cached. How the cache provider distributes the cache regions is implementation- or configuration-specific. I still think that this kind of cache architecture is questionable: I'd rather recommend setting appropriate limits on the cache size on each node rather than distributing data. The lookup is probably more expensive in any case than a database hit, at least for transactional data.