We've been experimenting with ways to store large collection of items in the cache and have not been able to settle on the right way to do this. One of the major challenges is that collections are aspectized in the pojo cache. We've found that collections that have hundreds of items begin to suffer performance degradation. When collections get to a thousand or more, CPU consumption almost takes over the entire system and it appears that all of the time is being spent weaving the collection. The alternate approaches we have taken are:

- Structure the tree cache such that items in the collection are all under a common node and then use the findAll capability to treat each item in a collection as an individual tree node. This eliminates collections all together but querying the entire collection is somewhat expensive
- Use a Java class that implements Serializable (not annotating Replicable) having a single property which is a collection. This essentially forces the cache to treat the collection as a whole object and eliminates the weaving problem. This has proven to be better performing than the other approaches. But, it too has performance issues. Persisting a collection of 1000 small items takes around a minute or so. Compared to a weaved collection though, the weaved collection is almost impossible to persist at that size.

So, my question to the JBoss Cache team is what is the recommendation to store collections of items in the pojo cache that could get into the hundreds or even thousands? Has there been any testing with respect to large collections of objects? Thanks in advance for any feedback.