A reference architecture for cooperative transaction processing systems

This report presents a reference architecture for cooperative transaction processing systems. The architecture was developed in the ESPRIT III basic research project TransCoop (EP8012). The reference architecture provides a general framework, which covers the architectural needs of three application domains, namely cooperative authoring, design for manufacturing, and workflow management. The architecture may have many instantiations in concrete cooperative transaction processing systems. Some components or functionalities can be pruned if they are not needed in a particular implementation of the architecture. The components of the architecture interact through protocols defined in terms of function calls. The embedded data flows comply with an object model common to all components. In practice, the functions can be implemented in many ways, as long as the external behaviour complies with the specifications. The protocols may span over a computer network, i.e. both distributed and centralized realizations are possible. The requirements analysis of the three application domains has pointed out that cooperative applications are developed iteratively. The development process is fairly isolated from the execution phase and results in a specification which may be instantiated in a run-time environment many times. Thus, the reference architecture consists of two major components, a specification environment and a run-time environment. The specification environment contains tools for editing application scenarios, verifying and simulating them, and compiling them into representations used by the run-time environment. The run-time environment has a transaction management support system as its central component. It has a homogeneous interface to agents and to an object storage interface mapper, through which it interacts with its environment. The transaction management support system supervises many concurrent cooperative scenario instances guaranteeing the transactional properties specified for them. An agent of the run-time environment may have the pure functionality of a client towards the transaction management support system, whereas in certain cooperative scenarios they may behave as servers towards it. There are agents representing end-users of the applications. Some agents represent remote sites, while others represent programs which either use - or are used by - the cooperative scenarios. The run-time environment can be replicated and the instances can be made to cooperate. The architecture is scalable.