The cullCache() method is called as soon as the cache is filled to 97%. First the expired entries are deleted and if this does not help (one may assume that it does not help as deleteExpiredEntries() is called for every cache hit) and then cache entries are removed. It helps a lot if one can see how often the method was called, a high call number indicates a small cache. An average value for the last 3 hours would also be very nice as most administrators don't check the gui regular (and notifying the admin via IM / pubsub/ email is not implemented).
Guus, please evaluate this for the wish list. I think its a good idea.
Greg Thomas did some effort in this area, assigning to him for his thoughts on potential inclusion.
cullCache() only applies to when using the default, non-clustered cache.
The equivalent is performed internally by the Hazelcast clustered cache - and I think the Coherence clustered cache can probably be ignored at this point in time.
Perhaps a more useful measure might be a "high-water mark" in terms of %age. If the high water mark is frequently close to 100%, then your cache is too small. I like the idea of a "window" - three hours or otherwise - but at the moment I can't think of of an easy way to implement it.