An analysis of edge fault tolerance in recursively decomposable regular networks

Abstract

Fault tolerance of interconnection networks is one of the major considerations in evaluating the reliability of large scale multiprocessor systems. In the paper, the reliability of a family of regular networks with respect to edge failures is investigated using four different fault tolerance measures. Two probabilistic measures, resilience and restricted resilience, are developed, used to evaluate disconnection likelihoods using two different failure models, and compared with corresponding deterministic measures. The network topologies chosen for the present study all have the recursive decomposition property, where larger networks can be decomposed into copies of smaller networks of the same topology. This family of graphs includes the k-ary n-cube, star and cube connected cycle graphs, which have optimal deterministic connectivities. The probabilistic fault tolerance measures, however, are found to depend on topological properties such as network size and degree. >