LIKELIHOOD RATIO SENSITIVITY ANALYSIS FOR MARKOVIAN MODELS OF HIGHLY DEPENDABLE SYSTEMS

This paper discusses the application of the likelihood ratio gradient estimator to simulations of large Markovian models of highly dependabl e systems. Extensive empirical work, as well as some mathematical anal ysis of small dependability models, suggests that (in this model setti ng) the gradient estimators are not significantly more noisy than the estimates of the performance measures themselves. The paper also discu sses implementation issues associated with likelihood ratio gradient e stimation, as well as some theoretical complements associated with app lication of the technique to continuous-time Markov chains.