Methodologies for meeting hard deadlines in industrial distributed real-time systems

Abstract

Four new approaches for coping with strict timing constraints as encountered in industrial process control applications are presented. For each node in a distributed system, an asymmetrical two-processor architecture capable of guaranteeing response times is employed. One processor in each node is dedicated to the kernel of a real-time operating system. Its three reaction levels are constructively described by outlining their functional units and control procedures. Contemporary real-time computers are unable to manipulate external parallel processes simultaneously, and their response times are generally unpredictable. It is shown that these problems can be solved by endowing the node computers with novel process peripherals. They work in parallel and perform I/O operations precisely at user-specified instants. To cope with a transient overload of a node resulting from an emergency situation, a fault tolerant scheme that handles overloads by degrading the system performance gracefully and predictably is used. The scheme is based on the concept of imprecise results and does not utilize load sharing, which is usually impossible in industrial process control environments.<>