Length-based blocking and local estimations in distributed simulation: a case study

Investigates the use of a preventive length-based blocking strategy associated with local estimations. A logical process must block until the length of its input queue becomes greater than a threshold L _min. By blocking one can reduce the number of time errors, in order to allow the local estimations to handle the remaining time errors more efficiently. We seek for an empirical optimal trade-off between the initial blocking length, the speedup relative to a conservative simulation, and the accuracy of the results. Then, automatic tuning is performed to adjust the initial blocking length to the desired time error ratio. We apply this method to simulate an ATM network, and the results show that the length-based blocking strategy succeeds in reducing the number of causality violations with a very small blocking overhead. Also, both accuracy and speedup can be achieved