TY - GEN

T1 - On the potential of semi-conservative look-ahead estimation in approximative distributed discrete event simulation

AU - Fu, Desheng

AU - Becker, Matthias

AU - Szczerbicka, Helena

PY - 2013

Y1 - 2013

N2 - One major problem of distributed discrete event simulation is the poor performance due to the huge overhead for maintaining the order of causality, so that the execution time cannot be reduced significantly compared to sequential simulation. This holds especially when the processes are tightly coupled and the look-ahead is very short. On the other hand, results of many simulations are obtained from a number of independent outputs, which are of stochastic nature and a small deviation of a limited amount of outputs is acceptable. Acceptance of such deviations in a controlled way could affect a trade-off between the simulation accuracy and the execution time. The goal of our investigation is to develop a methodology to handle the trade-off. In this paper, we propose a new way of distributed simulation with semi-conservative look-ahead estimation, where we accept causality errors to a certain and limited extent. In our approach, we consider a semi-conservative estimation allowing limited over-estimation. If the look-ahead is over-estimated, unsolved causality errors will be resolved by a very efficient recovery procedure at the expense of simulation errors. Results from a case study demonstrate that our approach is able to maximize the look-ahead with respect to the predefined error bounds and can reduce the execution time of many simulations. We do however also point out the limitations of the mechanism and the trend of our further investigation.

AB - One major problem of distributed discrete event simulation is the poor performance due to the huge overhead for maintaining the order of causality, so that the execution time cannot be reduced significantly compared to sequential simulation. This holds especially when the processes are tightly coupled and the look-ahead is very short. On the other hand, results of many simulations are obtained from a number of independent outputs, which are of stochastic nature and a small deviation of a limited amount of outputs is acceptable. Acceptance of such deviations in a controlled way could affect a trade-off between the simulation accuracy and the execution time. The goal of our investigation is to develop a methodology to handle the trade-off. In this paper, we propose a new way of distributed simulation with semi-conservative look-ahead estimation, where we accept causality errors to a certain and limited extent. In our approach, we consider a semi-conservative estimation allowing limited over-estimation. If the look-ahead is over-estimated, unsolved causality errors will be resolved by a very efficient recovery procedure at the expense of simulation errors. Results from a case study demonstrate that our approach is able to maximize the look-ahead with respect to the predefined error bounds and can reduce the execution time of many simulations. We do however also point out the limitations of the mechanism and the trend of our further investigation.

KW - Approximative simulation

KW - Discrete event simulation

KW - Distributed simulation

KW - Look-ahead estimation

UR - http://www.scopus.com/inward/record.url?scp=84880651613&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:84880651613

SN - 9781627482769

T3 - Simulation Series

SP - 72

EP - 79

BT - Summer Computer Simulation Conference and Work in Progress, SCSC 2013 and WIP 2013

T2 - Summer Computer Simulation Conference, SCSC 2013 and Work in Progress, WIP 2013, Part of the 2013 Summer Simulation Multiconference, SummerSim 2013

Y2 - 7 July 2013 through 10 July 2013

ER -