IPDPS'06 Proceedings of the 20th international conference on Parallel and distributed processing
| Hi-index | 0.00 |
In faster-than-real-time simulation (FRTS), a valid model may be maintained in real time through comparing simulation results with the corresponding system observations at regular time intervals. As, in the general case, simulation does not reach a steady state within each such interval, experiments must involve multiple replications, which must be completed (i.e. reachpredictions) within the given time. In the past, methods have been introduced for examining the possibility of executing a specific number of replications, based on measurements obtained from a number of initial replications. We discuss how such a method may be applied, both in sequential and distributed simulation, for determining the maximum number of replications thatmay be executed in the given time. Nevertheless, n may be less than a minimum nmin required to support the reliability of results. To deal with this case, we need to decrease the time period for which predictions are made, to save up execution time. We propose method extensions for determining the time interval for which results may be reached by nmin replications. Experimental results from various models are presented to substantiate the validity and effectiveness of the proposed method.