PROBABILISTIC DIAGNOSIS OF MULTIPROCESSOR SYSTEMS

Abstract

This paper critically surveys methods for the automated probabilistic diagnosis of large multiprocessor systems. In recent years, much of the work on system-level diagnosis has focused on probabilistic methods, which can diagnose intermittently faulty processing nodes and can be applied in general situations on general interconnection networks. The theory behind the probabilistic diagnosis methods is explained, and the various diagnosis algorithms are described in simple terms with the aid of a running example. The diagnosis methods are compared and analyzed, and a chart is produced, showing the comparative advantages of the various diagnosis algorithms on the basis of several factors important to probabilistic diagnosis.