The Two-Processor Reliability of Hierarchical Large-Scale Ring-Based Networks

In this paper we present the reliability analysis of ring-based hierarchical directed networks in the context of large-scale multiprocessors. A wide range of hierarchical structures based on the basic, dual counter-rotating, and braided (skip-a-node) rings are analyzed and compared in thispaper. We have adopted the two-processor (two-temzinal) reliability criteria for our analysis. The results of computing the system reliability for thousands of processors show that hierarchical structures constructed of counter-rotating and braided rings significantly improve the reliability of hierarchical structures constructed of basic rings. For most of the hierarchies the braided structures have better survivability andprovide higher reliability than the counterrotating structures for long mission times. Our results show that uniform hierarchical structures tend to provide better reliability than nonuniform structures. The results are of value in designing real t ime systems, large-scale architectures, and general computer networks. Because Scalable Coherent Interface (SCI) appears to be a promising interface for high speed interconnections, we use it to illustrate both the approach and the applicability of our results. The results apply also to similar ring architectures such as token rings, slotted rings, and other unidirectional ring networks.