News > Detail
View [ 1328 ]
Dr. Zeigler is invited to give a keynote speech at the Spring Simulation Multi-Conference, Alexandria, VA 2015 (http://www.scs.org/springsim)
Abstract of the keynote speech
Amdahl asserted that a program can run no faster than inversely related to the number of processors compared to what it can on a single processor. This speed up relation was originally stated as an assumption and has never been proved as such. For discrete event simulation, this implies that a distributed simulation is similarly limited in speed. In this paper, we derive a proof based on first principles specifically in the context of distributed simulation. The formulation and proof are simple yielding new insights into distributed simulation dependence on factors such as event work load distribution, numbers of processors, and inter-processor communication. We conclude with an interpretation of the theory to parallel DEVS simulation and show that on the average the standard DEVS distributed protocol may require at most 60% more run time of the best possible method conservative or optimistic method. Moreover, the DEVS protocol converges to the best possible speedup for models with moderate to high coupling. We conclude with a discussion of open research issues this approach raises.