Parallel and distributed framework for standalone Monte Carlo simulation using mapreduce

This paper deals with an efficient and robust parallel and distributed simulation framework for standalone Monte Carlo simulation based on a MapReduce computing framework. The Monte Carlo simulation method is inherently computing-intensive and requires many replicated simulation runs to get meaningful statistical results. Thus, it is important to reduce total simulation time by exploiting hardware and/or software as well as to reuse existing standalone simulation programs with little modification for the replicated simulations. To cope with this situation, we propose a general framework that turns a stand-alone Monte Carlo simulator into a chain of MapReduce jobs in order to run the simulation on a MapReduce framework such as Hadoop. A case study of an air defense simulation on 16-node Hadoop cluster illustrates that the proposed framework is feasible and fully utilizes the merit of the parallel and distributed computing environment.