The Goal:

Be able to program a system with multiple related objects purely from an object oriented perspective, yet have these objects magically persisted (and un-persisted).

General:
• Update When (Fine/Course Grained)
• Transactions (Begin, Commit, Rollback, and of what)
• Multi-Threading and Locking, Optimistic or Pessimistic (and when initiated)
Object References:
• Owned vs regular/simple (deleting?)
• Hard vs Soft (deferred instantiation)
• Collections (Views and Updates)
• Un-Persisting Initial Object Graph Reference(s)
• Persisted Object (Un-Persisting) Instantiation Process
• Reconstitution of Non-Persisted (singleton?) Referenced Objects
Scalability:
• Memory Usage - Complete Object Graph (DB) in Memory at Once
• Select/View Process - Query Language or Walk the Complete Object Graph
• (lite vs full views and/or partial vs full result sets)

Object Oriented Purity with Persistence of a Large Scale Object Graph is NOT a Reasonable Goal.

• If we can not achieve object oriented purity, can we at least achieve no vendor lock in?
• Can we achieve no vendor lock in & retain the simple Unit Testing of simple objects?

• Open Source (no vendor lock in)
• Change Perspective, instead of OO think TO (Table Oriented)