Limited Copies and Leased References for Distributed Persistent Objects

As businesses become global organisations and as e-commerce opens up markets to customers across the Internet, demand grows for increasingly ambitious distributed software applications and platforms. Where these applications run over potentially huge collections of data, sophisticated management of data storage and communication is required. There is a need for well-integrated persistence and distribution support that considers the implications for long-term maintenance of valuable persistent data. Orthogonal persistence is intended to ease the programmer's job by providing support for data management that is integrated with a programming language. The simplicity of the orthogonal persistence model argues for its use in distributed systems, in order to make life simpler for the application programmer. PJRMI is an implementation of Java RMI for the orthogonally-persistent PJama platform. This dissertation addresses two problem areas raised by combining orthogonal persistence with support for distributed applications. These problem areas are illustrated by PJRMI. The first problem is raised as a consequence of attempting to provide the illusion of a persistent connection between stores. Distribution-related errors easily break this illusion. In an open system, it can be difficult to determine when an object should become persistent by remote reachability. In the long term, persistent references to remote objects threaten the maintainability of the persistent stores involved. A solution has been implemented to address the problems raised by maintaining persistent references between distributed stores. Greater autonomy of individual stores is achieved by limiting remote access to objects to a duration of time associated with a specific distributed application's lifetime. Within the application's lifetime, the benefits are retained of persistence of inter-store references for resilience. The second problem is encountered when copying object graphs between stores. Large object graphs tend to build up in persistent stores over time. Copying such large object graphs can be prohibitively expensive in terms of resources and performance. A programmer may assume that the size of graph they are copying is acceptable, based on their knowledge of a system in its infancy. However, the problem is that, in a long-lived system, their assumptions may be challenged, since the size of an object graph and the context in which it is used are more likely to change during a persistent object graph's lifetime. The combination of a typically statically-defined policy for passing objects to remote sites and programmer assumptions that fail to take into account the lifetime of an object can also result in other problems. These problems include failure to support different requirements on remote use of the same object graph by different applications during that object graph's lifetime. A solution has been implemented to address the problems raised by remote copying of large object graphs. Flexibility of control over such copying is achieved. Separation of policy from object definition ensures flexibility. Choice of object-copying policy for a specific distributed application's lifetime provides control, while ensuring it is adaptable to changes in size of persistent object graphs over their lifetime and to changes in the context in which these graphs are used