Guaranteeing utility level of a firm real-time system using control theory

It is argued that time/utility functions (TUF) besides deadline and the utility accrual scheduling paradigm provide a more generalized and flexible approach to characterize real-time systems behavior. On the other hand, controlling the transient behavior of real-time systems is a quite important issue, whereas many such systems operate in unpredictable environments. Further, open-loop real-time scheduling algorithms, especially in overloaded conditions, do not satisfy the desired performance requirements. This paper proposes a closed-loop utility control algorithm for a firm real-time (FRT) system with a predetermined target for the accrued utility by the individual jobs who meet their deadline. In order to lead the accrued utility to the target, the algorithm uses DVS to dynamically adjust the processor speed. In the process of designing the algorithm, an established control theoretic method has been used. The method has been applied to the FRT system with three scheduling policies, namely, FCFS, EDF, and GUS. The efficiency of applying the proposed utility control algorithm to the system with these policies is investigated. Observations confirm that in spite of EDF, GUS is a quite suitable scheduling policy to get a relatively constant utility by the jobs in the system.