Efficient estimation of the mean time between failures in non-regenerative dependability models

In this paper we discuss fast simulation techniques for estimating the steady-state mean time between failures (MTBF) in non-Markovian models of highly dependable systems. The key is to use a ratio representation of the MTBF, in which the denominator is closely related to the probability of a rare event and is therefore amenable to estimation using importance sampling. A simulation methodology based on splitting and batch means, used for steady-state estimation in non-regenerative systems, can then be employed. Experiments using this methodology yield good results.