An Adaptive System-Level Diagnosis Approach for Mesh Connected Multiprocessors

Traditional adaptive centralized system diagnosis assumes a fully connected network topology, hence it can not be used in a number of classes of multiprocessor systems, such as meshes. This paper proposes an adaptive system-level diagnosis algorithm for meshes with wraparound, such as Intel Paragon machine. It is proved that the diagnosis cost required by the proposed approach is lower than the known diagnosis algorithms which can be applied to mesh architectures. Also over-d fault problem can be efficiently solved by our method, where d is the diagnosability.