RESTART vs Splitting: A comparative study

In this paper the comparative study is made for both transient and steady-state simulations. Both original methods and also their truncation versions with different depths of truncation are compared. The models used for the study include several Jackson and non-Jackson networks and also a general model of reliability with different types of components. The main conclusion of the study is that RESTART always behaves significantly better than Splitting. Another finding of this paper is that an additional gain (up to more than 50%) is achieved with RESTART with prolonged retrials of depths 1 or 2 in models where many thresholds can be set.