Probabilistic diagnosis algorithms tailored to system topology

Abstract

The authors previously (1989) presented algorithms in which if at least two processors perform tests on any given processor, the probability of correct diagnosis approaches one as N to infinity if the number of tests performed by each tester on each processor under test is O(log N). The algorithm was based on a comparison approach to probabilistic system-level fault diagnosis in which processors may perform multiple tests on other processors. Here they present a new hierarchical testing algorithm for this model and show that asymptotically efficient testing can be done when the product of number of testers*number of tests each performs on a processor grows as O(log N) as N to infinity . The method thus preserves the topological flexibility of the previous method, while allowing the number of tests each tester must perform to be tailored to the requirements of the topology.<>