Gianpaolo Ghiani, Pasquale Legato, Roberto Musmanno, Francesca Vocaturo / Computing, 2004, Vol. 3, Issue 3, 7-12 7 OPTIMIZATION VIA SIMULATION: SOLUTION CONCEPTS, ALGORITHMS, PARALLEL COMPUTING STRATEGIES AND COMMERCIAL SOFTWARE Gianpaolo Ghiani 1 , Pasquale Legato 2 , Roberto Musmanno 2 , Francesca Vocaturo 2 1) Dipartimento di Ingegneria dell’Innovazione Università degli Studi di Lecce, 73100 Lecce, Italy gianpaolo.ghiani@unile.it 2) Dipartimento di Elettronica, Informatica e Sistemistica Università della Calabria, 87030 Rende (CS), Italy {legato,musmanno, vocaturo}@unical.it Abstract. Simulation optimization (or optimization via simulation) is defined as the optimization of performance measures based on outputs from stochastic simulations. Although several articles on this topic have been published, the literature on optimization via simulation is still in its infancy. In this paper the research in this field is reviewed and some issues that have not received attention so far are highlighted. In particular, a survey of solution methodologies is presented, followed by a critical review of parallel computing strategies and commercial software packages. A particular emphasis is put on problems with discrete decision variables. Keywords: Stochastic Systems, Simulation Optimization, Metaheuristics, Parallelization Strategy 1. INTRODUCTION Several problems arising in complex system design and operations are characterized by a inherent stochastic nature. While analytical models are viable under relative easy assumptions, their use becomes more and more difficult as the complexity of the system increases. In such situations, we resort to simulation optimization modeling as an effective alternative. Simulation optimization (or optimization via simulation) is defined as the optimization of the performance measures of complex systems based on outputs from stochastic (primarily discrete-event) simulations (see Figure 1). More formally, the class of problems we are concerned amounts to finding a solution θ such that: ) ( min θ θ f Θ ∈ (1) where f(θ)=E[L(θ,ω)], L is the sample performance measure, θ∈Θ represents the (vector of) input variables (or controllable parameter settings), and ω represents a sample path (simulation replication). A particular setting of the variables is usually called either a configuration or a design while outputs are named performance measures, criteria, or responses. The constraint set Θ may be either explicitly given or implicitly defined. For simplicity in exposition, we assume throughout that the minimum exists and is finite, e.g., Θ is compact. Fig. 1 - Simulation-Optimization approach Stochastic discrete-event simulator Optimization subroutine candidate solution performance estimate computing@tanet.edu.te.ua www.tanet.edu.te.ua/computing ISSN 1727-6209 International Scientific Journal of Computing