A Dynamic Scheduling Algorithm for Real-Time Expert Systems

Computational characteristics of real-time expert systems have been the subject of research for more than a decade. The computation time required to complete inferences carried out by expert systems present high variability, which usually leads to severe under-utilization of resources when the design of the schedule of inferences is based on their worst computation times. Moreover, the event-based aperiodic activation of inferences increases the risk of transient overloads, as during critical conditions of the controlled or monitored environment the arrival rate of events increases. The dynamic scheduling algorithm presented in this article obtains statistical bounds of the time required to complete inferences on-line, and uses these bounds to schedule inferences achieving highly effective utilization of resources. In addition, this algorithm handles transient overloads using a robust approach. During overloads our algorithm completes nearly as many inferences as other dynamic scheduling algorithms, but shows significantly better effective utilization of resources.