Custom memory pool.

Question

Hi!
I have been studying different ways to implement object cache for server application(s). At the server there are XML files which have to be parsed and validated before delivering to clients. Clients are also sending XML data back to be validated and stored by the server. It would be nice to have object cache between servlets and IO to make server run faster. The ultimate requirement for cache is: There have to be way to set upper limit for caches memory footprint to avoid OutOfMemory.
I found couple of third parties object caches like Java Caching System (JCS), OSCache, Commons Collections, JCache API but none is supporting cache policy to fulfil my requirement. Am I right? Is there any others? (Open Source)
I started also desining my own implementation of object cache but the problem is: How to calculate object size in bytes.
I tried to solve problem by creating interface Cacheable that contains method sizeOf(). It is quite easy to implement and calculate approx size of (simple data) object except String. I wrote my on custom String class to get size. I also wrote some Cacheable implementations of collection classes too. But next problem was that (object) cache become too specific.
Okey, now JMX questions : )
1. Question is: Is it possible to use JMX to create totally new memory pools within same VM for object cache. Why? It would be easy to examine memory usage of cache by examining memory usage of those pools.
2. Is there way to set a classloader to load certain classes to those memory pools.
3. Is it possible to set a "rule" that instances of those classes are placed also into new pools.
4. Does JMX inspection have some severe performance drawbacks?
Plan B
I have also considered idea of running object cache as independent service in its own VM. And connecting to it by RMI. Any comments? Performance issues? Have anybody done this?
Thanks!

Answer 1

I do not think it is possible for user code to create a new memory pool in the sense of the ones that show up in JConsole's Memory tab ("PS Eden Space", "PS Perm Gen", and so on.)
I would in any case advise against getting into the details of object sizes. The question is very difficult - for example, is the size of an object the total size of all objects reachable from it? If so then if x and y are two objects that have part of their object graph in common, adding the sizes you compute for x and y will produce a size that is too big. If you want to get involved in this anyway, a good starting point is <http://www.javaworld.com/javaworld/javaqa/2003-12/02-qa-1226-sizeof.html>.
You can in fact get a number representing the size of a Java object by writing a premain method and then using Instrumentation.getObjectSize. See java.lang.instrument. But this size does not include objects referenced from the given object, so it does not really tell you how much memory the object is taking up. For example, all String objects will show the same size, because the char[] containing the actual contents of the String is another object that is referenced by the String.
The usual way to handle object caching is with SoftReference, see java.lang.ref. Admittedly, this does not give you an easy way to balance the sizes of different pools.