Diagnosing arbitrarily connected parallel computers with high probability

A practical model for probabilistic fault diagnosis is presented. Unlike PMC-based models, the model allows testers to conduct multiple tests on the same processor. This allows the design of efficient probabilistic diagnosis algorithms with good asymptotic behavior, with minimal constraints on the connection structure of the multiprocessor system, in contrast to other deterministic and probabilistic approaches. In practical cases, the number of immediate neighbors of any processor need be no greater than two, which implies that the algorithm can be applied to any practical homogeneous parallel architecture. It is also shown how to make efficient use of tests by allowing the number of testing processors, and the number of tests performed by a processor to be traded off in achieving asymptotically accurate diagnosis