Meta-level architecture support for distributed objects

Abstract

The creation of distributed applications is often hindered by a lack of a priori knowledge of distributed object behaviour. Designers and builders of distributed systems can benefit from an environment which allows them to explore and experiment with various computational and structural models for application objects. Our ability to do this is limited by traditional distributed systems' tendency to mix domain-specific object description and distributed behaviour specification. By using an explicit meta-level architecture, we can transparently add a wide variety of distributed behaviours to objects with little effect on their base-level behaviour or code. We outline such an architecture and detail the design of various distribution mechanisms and policies (e.g. marshaling and replication). These are shown to be independent of base-object behaviour to such a degree that they can be recursively applied to the architecture in which they are implemented.<>