I'm are considering using a very large 2nd level cache for a large app with many users. Since there are so many sessions and a large part of the DB would be in the 2nd level cache, it is a pity that each object used in a session is always copied. It would cause us to hit physical memory limit (30GB RAM).

Now I'm wondering: is it not possible for the session cache to use objects directly from the 2nd level cache and only create copies on write? That would be great for scaleability w.r.t. number of user sessions.

No, not out-of-the-box. This would also imply that your complete second-level cache is fully transactional and not give you much performance benefit. Session size certainly is an issue, but are you really sure that this affects you? Do you have benchmarks that show that this is a problem?

No, we're just at the start of a large project, so no experience/benchmarks yet.

We do have the amount of RAM to hold the whole active part of the DB in memory, and reads outnumber writes by a factor of 10 (estimated) or more. So a large cache should be of great benefit. However, individual user sessions may use a significant amount of total RAM (say 10%), so already 10 parallel user sessions would double the amount of used RAM simply by creating copies of objects that are not likely to be modified. That is quite a waste; it would be very nice if it could be avoided. Not only because of RAM usage, but also because of the strain it puts on the garbage collector.

The alternative would be to use one session for the small editing stuff, and use other session with flush/clear (or even bypass hibernate alltogether) for producing large lists of objects. Do you think the latter is the only way?

Thanks for any insight.

The Hibernate Session is _not_ the same as a user session.

I know, I was talking about 10 parallel hibernate sessions :). Typically we'll have 3000 or more parallel user sessions. With 10 real transactions parallel in progress, the amount of memory waste due to all the copying for (mostly) read-only objects is quite large.

Thats nothing, people have hundreds of parallel Hibernate sessions.