Commentary—Will Parallel Simulation Research Survive?

Abstract

In his feature article, Fujimoto pointed out that the success of parallel simulation depends not only on the speedup that can be achieved through parallel simulation but also on the effort required to develop the parallel code. The general simulation community will only recognize parallel simulation technology if several practical simulation applications are found such that the total cost of the code development and code execution for parallel simulation is much less than the cost of sequential simulation. The sequential execution times of most examples listed in Table II in the feature article were less than two hours. It is difficult to justify parallel simulation as the right approach to follow if one needs to spend one additional day (or more) to develop parallel code in order to reduce a two hour execution time to several minutes. Most parallel simulation research focuses on reducing execution time of parallel simulation. Fujimoto proposed several approaches to address the parallel code development issue. I will consider this issue in a different aspect as a supplement. In addition, I will discuss how to predict the performance of parallel simulation, an issue not explicitly discussed in the feature article. Another issue not addressed in the feature article is the distributed computing platform for parallel simulation. Many examples listed in Table II in the feature article were run on shared memory architectures, and none of the examples were run on a network of workstations, a widely used hardware platform which cannot be ignored by parallel simulation researchers. It is difficult to gain acceptable speedup for parallel simulation in a network of workstations due to long communication delays. One possible solution is to connect the workstations using a high speed network. Another solution is to carefully map the processes to the processors such that inter-processor communications are reduced. How to do the latter efficiently is still an open issue. INFORMS Journal on Computing, ISSN 1091-9856, was published as ORSA Journal on Computing from 1989 to 1995 under ISSN 0899-1499.