Towards Credible and Fast Quantitative Stochastic Simulation

摘要

The computer revolution initiated in the second half of the twentieth century has resulted in the adoption of computer simulation as the most popular paradigm of scientific investigations. It has become the most commonly used tool in performance evaluation studies of various complex dynamic stochastic systems. Such reliance on simulation studies raises the question of credibility of the results they yield. Unfortunately, there is evidence that many reported simulation results cannot be considered as credible. In this paper, having briefly overviewed the main necessary conditions of any trustworthy simulation study, we will focus on the issue of sequential on-line analysis of simulation output data, the most practical way for obtaining statistically accurate results from simulation studies. The perils and pitfalls of quantitative sequential simulation will be considered, together with its fast distributed version, based on concurrent execution of simulation on multiple processors and known as Multiple Replications in Parallel (MRIP). We will discuss main properties and limitations of MRIP, as well as its implementation in Akaroa2, a simulation controller designed at University of Canterbury in Christchurch, New Zealand, automatically executing MRIP on clusters of computers in local area networks.