OutOfMemory from UnitOfWork and strange Merge behavior.

Question

My object model is like this:
Object A has a list of object B(One-to-many mapping, readOnly for this attributes).
Each object B back refers to A(bi-directional, one-to-one mapping).
My application has only one DatabaseSession.
In one thread, I need to go through a loop for about 200 to 500 times.
The code is like this
int n=200;
for (int i=0; i<n; i++)
     UnitOfWork uow=session.acquireUnitOfWork();
     A aclone=(A)uow.registerExistingObject(a);
     B b=....;
     aclone.addB(b); // b's A will set to a inside of this method.
     long startTime=System.currentTimeMillis();
     uow.commit();
     long endTime=System.currentTimeMillis();
     System.out.println(String.valueOf(endTime-startTime));
If n is less than 100, the topLink behaves consistently.
It takes less than 100 milliseconds to add an object B into database.
However if the n is 200 or bigger, the time for each UnitOfWork.commit() is sharply
increased(from 1000 to 8000 milliseconds). Finally it throwed an OutOfMemory exception.
I have a DescriptorEventListener registered for my class. This DescriptorEventListener monitors
the postMerge event. I expected that in each loop, the merge of the clone to
original object should be called for the one new object I inserted. It is indeed called
for the object b. However The merge is called for all existing bs
associated with A. Maybe it is why it takes more and more times to add a new B to A atomically.
Anyone has the same experience or any suggestion to solve this problem.

Answer 1

When I added the Nth b to A, A only has (n-1) b associated to it (the bs I added in the loop).
I tried your suggestion. It did not work.
I got this exception.
EXCEPTION [TOPLINK-6004] (TopLink - 9.0.3 (Build 423)): oracle.toplink.exceptions.QueryException
EXCEPTION DESCRIPTION: The object [com.test.A@7c4c51], of class [class com.test.A], with identity hashcode (System.identityHashCode()) [8,146,001],
is not from this UnitOfWork object space, but the parent session's. The object was never registered in this UnitOfWork,
but read from the parent session and related to an object registered in the UnitOfWork. Ensure that you are correctly
registering your objects. If you are still having problems, you can use the UnitOfWork.validateObjectSpace() method to
help debug where the error occurred. For more information, see the manual or FAQ.
QUERY: WriteObjectQuery(com.test.Parent1@7c4c51)
     at oracle.toplink.exceptions.QueryException.backupCloneIsOriginalFromParent(Unknown Source)
     at oracle.toplink.publicinterface.UnitOfWork.getBackupClone(Unknown Source)
     at oracle.toplink.publicinterface.UnitOfWork.getBackupCloneForCommit(Unknown Source)
     at oracle.toplink.queryframework.ObjectLevelModifyQuery.prepareForExecution(Unknown Source)
     at oracle.toplink.queryframework.WriteObjectQuery.prepareForExecution(Unknown Source)
     at oracle.toplink.queryframework.DatabaseQuery.execute(Unknown Source)
     at oracle.toplink.publicinterface.Session.internalExecuteQuery(Unknown Source)
     at oracle.toplink.publicinterface.UnitOfWork.internalExecuteQuery(Unknown Source)
     at oracle.toplink.publicinterface.Session.executeQuery(Unknown Source)
     at oracle.toplink.publicinterface.Session.executeQuery(Unknown Source)
     at oracle.toplink.internal.sessions.CommitManager.commitAllObjects(Unknown Source)
     at oracle.toplink.publicinterface.Session.writeAllObjects(Unknown Source)
     at oracle.toplink.publicinterface.UnitOfWork.commitToDatabase(Unknown Source)
     at oracle.toplink.publicinterface.UnitOfWork.commitRootUnitOfWork(Unknown Source)
     at oracle.toplink.publicinterface.UnitOfWork.commit(Unknown Source)
     at TestAdd.main(TestAdd.java:58)
That means that I have to register the object A. This is really bad feature for UnitOfWork. I just want to add a new Object B. Why does I need the burden from A?
This brings up another question. If I do not want the burden from A, I need to break the association from A and B and maintains the relationship manually. However I really wish my persistence layer can handle the association transparently for me.
I also notice that you suggest all the 200 bs in a UnitOfWork. In my situation, I need 200 UnitOfWork, I need to add each b in a separate UnitOfWork. When I call the UnitOfWork.commit(), the UnitOfWork is released cleanly. Is it right?
Thansk