Dependability evaluation of hierarchical systems

Hierarchical design represents a natural solution for managing the growing complexity of computing systems. This leads to systems that are hierarchically structured from a logical or physical viewpoint. Because of the inherent complexity of such systems, the incorporation of fault-tolerance features is highly recommended, to achieve the required dependability level. Hence, it is important to develop evaluation tools that allow to analyse the effectiveness of different fault-tolerance mechanisms. We present a systematic procedure for the construction and evaluation of Markov models for transient dependability analysis, which fully exploits the hierarchical structure of the system under evaluation. A thorough time/space complexity analysis of the proposed procedure shows that it allows a considerable time and space saving with respect to a naive approach to the same problem.Work partially supported by CNR, Progetto Finalizzato Sistemi Informatici e Calcolo Parallelo, under grant no. 91.00905 PF96, and by MURST, Progetto 40% Performability di sistemi distributi e paralleli