STATISTICAL METHODOLOGY FOR WEB-BASED SIMULATION William E. Biles Jack P. C. Kleijnen Department of Industrial Engineering Department of Information Systems University of Louisville Tilburg University 304 J. B. Speed Building P. O. Box 90153 Louisville, KY 40292 USA 5000 E Tilburg, The Netherlands ABSTRACT This paper describes a procedure for assigning simulation trials to a set of parallel processors for the purpose of conducting a simulation study, involving a complex simulation model, in near real time. Unlike distributed simulation, where a complex simulation model is decomposed and its parts run in a parallel environment, the parallel replications approach discussed here involves running simulation replications to completion for the entire model. The unique element here is that the workload involved in running the simulation study is too time consuming to execute on a single workstation, so that the simulation analyst must utilize computer resources available through the web. New statistical methodology is needed for running a complex simulation study in a web- based, parallel replications environment. INTRODUCTION Biles and Kleijnen [1999] described a Java-based approach for allocating a set of K = RS simulation trials to P parallel processors available through the world-wide web, where there were R replications of the simlation at each of S sets of input conditions. The methodology proposed there assumed that the simulation model had N input variables X i , i = 1,…,N and M system responses Y j , j =1,…,M, and that the objective of the simulation effort was to carry out the predictive phase (see Kelton, Sadowski and Sadowski [1998]) of a simulation study using experimental design, response surface methodology, or an optimization approach. The key component of the methodology described by Biles and Kleijnen [1999] was a program called the Simulation Manager that resided on a central processor under the control of the simulation analyst. The Simulation Manager allocated the K simulation trials to the P processors, sent a file to each processor specifying the input parameters necessary for that processor to carry out its assigned workload, and received back a file at the completion of that assignment that gave the summary statistics for the simulation activity. Biles and Kleijnen [1999] assumed that the Simulation Manager would assign a simulation workload only to idle processors according to a priority scheme in which the faster processors among the P slave processors received a proportionally higher workload than the slower processors. But later on, Biles, Marr, Storey and Kleijnen [2000] showed that a simulation task assigned to the p th processor was executed even if that processor