The solution of optimization problems in real world applications usually involves an enormous amount of computing time in which the use of parallel computers is of great value. The advent of high performance computer systems has led to substantial time savings in solution processes in operations research (OR). Moreover, large-sized problems have become tractable to a much higher extent. The goal of this special issue on high performance computing in operations research is to provide the reader with state-of-the-art issues in the intersection of parallel algorithm design and mathematical programming. The results presented here clearly show that parallel methods for OR are maturing into sophisticated tools and theoretical frameworks. Nevertheless, we note that the eort to harness the potential power of parallelism should not be underestimated. The existing techniques are not readily available, and hence much work remains to be done.
The issue is divided into two parts, composed of four papers each. The ยฎrst part is application driven and tackles several problems arising in ยฎnance, scheduling, and location. The ยฎrst paper, by Pยฏug and Swie รtanowski, surveys stochastic dynamic problems from the ยฎeld of ยฎnancial planning, an area steadily growing in importance. Then, Bourbeau, Crainic and Gendron compare a number of load balancing schemes for parallel branch-and-bound applied to location problems. The last two papers in this section deal with task scheduling in multiprocessors. This well-deยฎned sub-class of scheduling problems in OR happens to be one of the key issues for successful use of parallel hardware. Two approaches are shown here to solve such problems. One, by Corr ea, uses a branch-and-bound algorithm to obtain approximate solutions by fathoming nodes that cannot produce signiยฎcantly better solutions, whereas Porto, Kitajima and Ribeiro show the eectiveness of a parallel approximative algorithm based on tabu search for this same class of problems.
The second part of the issue is more of an explorative nature. Its four papers propose theoretical frameworks for the parallelization of well-established sequential OR techniques, namely tabu search, simulated annealing, dynamic programming, and linear programming. All the proposals are backed by extensive testing, either by simulation or by implementations in existing parallel systems. The paper by Toulouse, Crainic and Thulasiraman analyzes the eect of cooperatively searching the same solution space in dierent processors using tabu search. Next, Morales, Almeida, Rodr รตguez, Roda, Coloma and Delgado investigate the parallelization of dynamic programming from a cellular automata point of view. This part is