Optimal Arbitration of Control Tasks by Job Skipping in Cyber-Physical Systems

We consider cyber-physical systems where several physical plants are controlled by digital controllers implemented as periodic tasks executing in one embedded computer. If overload situations occur, control performances of the controlled plants may degrade seriously. We introduce an arbiter which accepts or drops jobs released by the tasks to avoid the overload situations. We consider a cost function to evaluate degradation of control performances by the drop. We assume that jobs are scheduled by the EDF algorithm and propose an arbitration algorithm to minimize the cost function so that all accepted jobs meet their deadlines and all plants are stabilized.