Parallel skeletons for Tabu search method

Abstract

We present two generic parallel skeletons for the tabu search method-a well known meta-heuristic for approximately solving combinatorial optimization problems. The first skeleton is based on independent runs while the second in the classical master-slave model. Our starting point is the design and implementation of a sequential skeleton that is used later as basis for the two parallel skeletons. Both skeletons provide the user with the following: a permit to obtain parallel implementations of the tabu search method for concrete combinatorial optimization problems from existing sequential implementations; there is no need for the user to know either parallel programming or communication libraries; and the parallel implementation of tabu search for a concrete problem is obtained automatically from a sequential implementation of tabu search for the problem. The skeletons, however, require from the user a sequential instantiation of the tabu search method for the problem at hand. The skeletons are implemented in C++ using MPI as the communication library and offer genericity, flexibility, component reuse, robustness and time savings. We have instantiated the two skeletons for the 0-1 multidimensional knapsack problem, among others, for which we report computational results.