Dependability Research Of Distributed System Based On Stochastic Petri Net

Dependability has come into being a matured theory system. It is the measurement indeces of quality of service of distributed systems. Stochastic Petri net (SPN) is an ideal analyzing tool in dynamic concurrence,asynchronism,nondeterminism. It is especially fit to model system and analyze dependability of system. Cluster system has become an infrastructure, which constructs high performance distributed systems. The quality of service (QoS) offering for users will be affected by dependability at large. Standby system is the one of policies applied to increase dependability of distributed systems. It can obtain higher dependability by increasing redundancy resource, and still maintain normal functions when distributed systems failures occur.This paper is mainly about dependability of distributed systems based on SPN. The author takes advantage of SPN to model front-end dispatcher and backend workers respectively, and presents SPN model of the whole cluster system. The necessary repairing assurance is taken into consideration when modeling the cluster system. A set of dependable parameters of the cluster system are presented on the base of model, and the model is simulated to validate.The simulation results show that the stable availability of system will increase remarkably when the key system front-end dispatcher has been given necessary repairing assurance. The author introduces the concepts of standby multiprocessor systems and SPN expressions of cold standby system and warm standby system, and models the standby multiprocessor system by taking advantage of SPN, and futher presents different dependable parameters of standby multiprocessor system based on the SPN model analysis. The author gives out that if model designer try modeling system by Markov model, the structures of Markov chain will change largely when the change of system resources (for example the number of processors) occurres. If the system is modeled by SPN, it will obtain corresponding model by changing the number of tokens which denote resource in the place. Hence it will promote modeling efficiency and analysis ability of standby systems by bringing in SPN to dependability area.