Hierarchical modeling and dependability evaluation of distributed systems

We propose a new dependability evaluation method for a distributed system where programs and data files are replicated and allocated to computing nodes of the system in a redundant manner. The proposed method consists of two-level hierarchical procedures. At the lower level, the behavior of system components is analyzed using Markov models, while at the upper level, the whole system is modeled by a fault tree. The advantages of the proposed method are summarized as follows: (1) behavioral features of the system, such as coverage factor and global repair, are explicitly taken into consideration, and (2) dependability evaluation can be performed without causing state explosion, since Markov modeling is localized only to each component. Through extensive simulation studies, we show that the proposed method can accomplish exact evaluation with sufficiently admissible running times even for large distributed systems that previous methods cannot handle.