The most common approach is based on Corba like technology. In this approach, each component declares the services it provides. The process itself is executed by a special PSS which asks the “right” PSS to execute the “right” service when required. It is control based interoperability. We show that this kind of technology is suited as long as the process model can be partitioned in such a way that each partition (called a component) can be handled entirely by a single PSS.
The other basic approach is based on the availability of the state of a common virtual process. Each PSS can see the state of the common process, can change it and react to changes performed on that state by other components. Each component collaborates freely to the federation behaviour. It is state based interoperability. We show that this approach is flexible but it does not provide control of the process.
It is shown how these dual approaches can be reconciled, keeping the best of both paradigms. We claim that the choice of a paradigm or a combination must be dictated by the kind of interoperability required, not by platform technological limitations. This paper also claims that we need a clear distinction between the functional and operational aspects of a federation.
We have designed and implemented the APEL interoperability platform which supports both paradigms and their combination. This paper presents the architecture of that platform, and discusses its major characteristics. We discuss the experience gained and the issues still to be addressed.