Proceedings of the 2009 Industrial Engineering Research Conference Metaheuristics for Quadratic Assignment Problem (QAP) Thashika D. Rupasinghe and Mary E. Kurz, Department of Industrial Engineering, College of Science and Engineering, Clemson University, Clemson, SC 29634-0920, USA Abstract The Quadratic Assignment Problem (QAP), one of the difficult problems in the NP-hard class, is applicable to model many real world problems. Specifically, many combinatorial optimization problems are formulated as QAPs. Recent trends have been to use metaheuristics rather than exact methods to resolve QAPs due to the complexity and computational resource constraints. In literature many researchers have demonstrated the effectiveness of metaheuristics to solve complex real world problems. This study emphasizes on comparing the performance of tabu search and simulated annealing methods in solving QAPs. Furthermore, utilizes a large scale computational study to carry out an extensive parameter tuning process using the computational resources of grid computing. Keywords Combinatorial optimization, High-throughput computing, Metaheuristics, Quadratic Assignment Problem 1. Introduction The overall objective of the study is to explore quadratic assignment problem (QAP) as a test bed to develop robust metaheuristic designs utilizing the computation resources of grid computing. QAP is computationally difficult to solve, making it an ideal candidate for metaheuristic approaches. Although many researchers have investigated the QAP using pure and hybrid metaheuristics, very few have explored the effect of design parameters and parameter tuning on the performance of a particular algorithm using large scale computational resources. The authors have used QAP as the problem domain and developed two instances of commonly applied metaheuristics, tabu search and simulated annealing to conduct an extensive parameter tuning procedure. The findings of this study will contribute to the field of metaheuristic designs, development and computational studies in general. 1.1. The Quadratic Assignment Problem (QAP) The Quadratic Assignment Problem (QAP) was first stated as a mathematical model in 1957 [8] and many practical problems[2], [8] can be modeled as QAP including production line scheduling, assignment of gates to airplanes in airports, backboard wiring problems in electronics, campus and hospital layouts, typewriter keyboard designs, turbine runner balancing problems, processor-to-processor assignment in a distributed processing environment and many others. On account of its diverse applications, theoretical importance and its complexity, QAP has been researched by many researchers around the world [8]. The mathematical formulation of the QAP takes different forms, including integer linear programming (IP) formulation, mixed integer linear programming formulation (MIP), formulation by permutations, tree formulation and graph formulations. The most general way of formulating it uses the IP format and may be described within a context of a facility location problem. The objective is to assign facilities to locations in such a way that each facility is located in exactly one location and vice-versa. These decisions are represented by the decision variables ik x which take on the value 1 when facility i is located in location k and 0 otherwise. There are two data matrices associated with the problem: the distances, kp d between locations k and p and the demand flows, ij f , between facilities i and j. The facilities are allocated such that the sums of all possible distance-flow products are minimized [2]. A standard integer programming formulation follows: , 1 , 1 min n n ij kp ik jp ij kp fd xx    (1)