Modeling and integration of fault tolerance techniques in distributed systems
Y.-B. Shieh · 1990
Fault tolerance is an essential requirement in the design of a reliable system. The research addresses some fault-tolerant issues pertaining to hierarchical distributed systems and proposes a hierarchical model for supporting fault tolerance in such systems. A hierarchical system is a particular type of distributed system where each level has different characteristics and different fault-tolerant requirements. This implies that different fault-tolerant schemes are more appropriate for the various levels in the hierarchical systems. This dissertation examines the performance and fault-tolerant trade-off issues for fault-tolerant schemes. Further, the effect of the integration of the fault-tolerant schemes at the various levels is also addressed. In this dissertation, the Stochastic Petri Nets (SPN) are used as the primary tool to investigate the performance and fault-tolerant issues. Towards this end, the basic SPN is augmented by parameterized subnet primitives to incorporate the fault-tolerant schemes. Furthermore, a methodology for generating the SPN models for the various systems with the fault-tolerant schemes has been developed. This methodology can be easily represented by a set of algorithms which provide a user friendly tool to automatically generate the SPN model for complex systems. In some cases, since the state space of complex systems becomes very large, SPNs have been used in conjunction with probabilistic models. Based on these analytic models, we have investigated a number of issues of the fault-tolerant schemes at the various levels of the hierarchical system. In this regard, we considered each level in our hierarchical system to consist of a set of concurrent processes with synchronous and asynchronous communication. Both process failure and node failure are incorporated into the models. Different strategies based on centralized and distributed fault-tolerant schemes have been investigated both with respect to the inherent characteristics of the various levels in the hierarchical systems and their effect on the performance. Finally, the effect of the integration of the fault-tolerant schemes of the various levels on the overall performance has been investigated. This research provides a methodology for designing integrated fault-tolerant schemes for hierarchical distributed systems satisfying certain performance criteria.