Monte Carlo Simulation of Computer System Availability/Reliability Models

In recent years, there has been an increased interest in evaluating the availability and reliability of computer systems. For this purpose, we have developed a state of the art modeling package called SAVE (System Availability Estimator). This package has refined numerical matrix methods for computing the dependability measures of interest. These methods do, however, have their limitations since the growth of the size of the state space of these models is exponential in the number of component types. In this paper, we investigate the use of Monte Carlo simulation as an alternative for solving models with a large number of components. Since probability of failure is relatively small, it has been observed by many researchers that simulation takes a large amount of computational time. We show that the Importance Sampling variance reduction technique may be applied to reduce the simulation run lengths substantially. We give examples which demonstrate the viability of using simulation for the evaluation of large availability/reliability models. Finally, we describe a simulation program which has been developed for the analysis of models that can be generated using the SAVE modeling language. This program makes automatic use of the Importance Sampling technique to reduce the run lengths.