Reactive scheduling for hard real time systems

Constraint maintenance is a well-known and an effective general problem-solving strategy in AI. This approach has been successful in developing a reactive scheduler for VLSI plant operation. A constraint net models the real world problem and this net can be updated at little cost with an immediate check on the feasibility of the schedule. The favourable attributes of this approach for application in real time systems are its flexibility in handling dynamic scheduling aspects, minimal run time cost in establishing revised schedules, the constant availability of a feasible schedule for immediate use and the maintenance of the whole set of solutions. The paper discusses the practical issues both in relation to application of the technique in real time control and its implementation in hard real time systems. The application of the technique to real-time control focusses on the relationship of the primitives of the scheduler to typical high-level real-time constraints, as well as the extent to which the technique may be applicable. The paper also examines the incorporation of the scheduler in a real-time embedded system. When an infeasible schedule occurs a rule-based strategy is invoked to recover the feasible one. This implies the selective shedding of scheduled tasks with the probable consequences of an orderly shutdown of the affected area of the plant by the real-time system