On the Selection of Architectures for Distributed Computer Systems in Real-Time Applications

To achieve an acceptable balance between cost and performance in designing distributed systems, architectures must be selected to match the specific requirements of each application. Because of the long lead time in large projects like nuclear power plants, a decision to exploit distributed systems cannot wait until all pertinent information becomes available. Using established concepts, a number of "building blocks" are defined as necessary components of a representation from which system characteristics can be derived. A generalized design sequence is formulated to help identify three major problem areas: processing clusters, data communications and data base. The design of a distributed data base is the area of most concern as it is the least understood and is complicated by its interaction with the selection and implementation of a partitioning strategy. A representation of a generalized data base is built up from traditional structures and a concept of common ports is introduced to satisfy all information transfer requirements between local data bases. It is concluded that a design methodology for distributed systems can be built, through derivation and generalization, from techniques established for centralized systems and computer networks.