Dynamic scheduling for networked control systems

Abstract

An integrated approach, embracing both control and scheduling theories, is proposed to implement multiple control loops upon shared network and computational resources, where the network may additionally introduce packet losses. Each control system is first analyzed from a control-theoretic perspective in order to determine the asymptotic rate at which control signals must be computed to maintain stability and optimal performance despite network losses. Since required completion rates for control tasks are asymptotic, and network packet drops uncertain, the problem of scheduling multiple such control tasks upon shared computational resources does not map to known problems in real-time scheduling. It is therefore formalized here as a new form of periodic task scheduling problem -- one in which each task has an associated asymptotic completion rate requirement. Sufficient schedulability conditions are derived, and a dynamic scheduling algorithm designed, for solving such scheduling problems. This integrated methodology thus provides an effective way to incorporate network loss in the design of cyber-physical systems over integrated architectures. The use of this methodology is illustrated, and its efficacy demonstrated, upon an example system of five inverted pendulums.