Partitioning of even networks for improved diagnosability (multiprocessor systems)

A partitioning property of even interconnection networks is presented. This property is based on the Hadamard matrix combinatorial structure which is used to partition the network into identical spheres. A semidistributed diagnosis algorithm helps to identify faulty components (nodes/links) in each sphere. Using the diagnosis result of each sphere on a global level, the overall diagnosability or even networks is improved drastically over previous results. This partitioning and diagnosis scheme can be used for almost any network that has some group-theoretic representation, including the binary hypercube.<>