Construction and analysis of fault-secure multiprocessor schedules

Issues involved in the design and analysis of fault-secure schedules for multiprocessor systems are investigated. A formal characterization of fault-security for a single fault is developed and generalized for multiple faults. The single fault characterization is used in the construction of fault-secure schedules for several classes of computation trees. The schemes produce schedules that are either shorter than or use fewer processors than the schedules produced by currently known methods. Further, lower bounds on schedule lengths are developed to prove that the schedules produced by the schemes are optimal or close to optimal. The characterization for multiple faults leads to an efficient algorithm to determine whether a given schedule is fault-secure for any fixed number of faults. It is shown that when the number of faults is not fixed, the problem of determining whether a schedule is fault-secure is computationally intractable.<>