Fault Injection Simulation: A Variance Reduction Technique For Studying The Performance Consequences Of Rarely Occurring Failures In Communication Networks

In this short paper a new technique, called Fault Injection Simulation (FIS), is discussed that is suited for deriving results for steady-state measures in discrete event dynamic systems which are strongly influenced by rarely occurring events. FIS is based on division of the observations in those that are affected and those that are not affected by these rare events. If methods are available FIS can be used as a (partly) analytical technique, else as a pure fast simulation technique. Under intuitively appealing assumptions FIS gives an unbiased estimator and a variance reduction. In this short paper we discuss FIS from a practical point of view; when can FIS be used and how should FIS be used. Furthermore, we show how to adjust FIS during the simulation to benefit maximally from its variance reduction capabilities.