Improving the execution of groups of simulations on a cluster of workstations and its application to storage area networks

Parallel simulation methods can be used to reduce the execution time of simulations of complex systems. This approach is being used to improve the execution time of a storage area network (SAN) simulator designed in our department. From our experience in planning simulation experiments, we have realized that, in most cases, a simulation experiment (group of simulations) is executed while varying only one input variable, which usually corresponds to the input, workload or a configuration model parameter. We propose two methods to reduce the overall execution time of a simulation experiment using a cluster of workstations. The first method uses the first simulation in order to tune the rest of the remaining work to be done in the experiment. The second method, based in the first one, tries to minimize the negative influence of the initial transient period by chaining the simulations in the experiment. We show that these two methods noticeably decrease the overall execution time needed to run the simulations that compose the experiment.