Local Diagnosis in Massively Parallel Systems

A new method for local diagnosis in regularly interconnected massively parallel systems with fault-clusters presented here is based on comparison of self-test results and examination of the local neighbourhood. Several local diagnostic procedures can be initiated concurrently by system units. The radius of the neighbourhood to be examined depends on the size of the neighbourhood to be diagnosed, the maximal size of fault-clusters, and the minimal distance between fault-clusters. The local diagnostic procedure and the upper bound for the examination radius are given. Keywords: Local diagnosis; Diagnosability; Examination; Surrrounding; Massively parallel system; Fault-cluster; 1 Introduction A massively parallel system with regular interconnection topology and replicated system units is a promising instrument for increasing the computational power of computers. Most existing algorithms for such systems operate on local data and require that the system be of regular structure. A popular..