Including Selective Fault - Tolerance inReal - Time Multiprocessor Schedules 1

An increasing number of applications are demanding real-time performance from their multiprocessor systems. For many of these applications, a failure may produce disastrous results. However, the area of fault-tolerant hard real-time multiprocessor scheduling has scarcely been investigated. Consider fault-tolerant static scheduling of tasks of varying degrees of criticality on a multiprocessor system constrained by precedence and hard real-time requirements. We have developed several algorithms that attempt to selectively build as much fault-tolerance as possible into an existing schedule while observing precedence and hard real-time constraints. Our algorithms give critical tasks priority when scheduling backups to be used as a method of fault-tolerance. Two types of backups, replicas and deallocating backups, are studied, and the practice of backup overloading is employed in our algorithms. An interactive graphical user interface was created to run our algorithms and display their resultant schedules, along with some statistical information about each schedule. Our results show that selective fault-tolerance may be built into schedules by increasing processor utilization without compromising the integrity of the original schedule. Furthermore, simulation results provide valuable information about backup characteristics which may be used to aid in algorithmic decision-making. 1