Parallel branch-and-bound skeletons: Message passing and shared memory implementations

Abstract

The aim of this article is to describe the design and the implementation of skeletons for the Branch-and-Bound technique. The library developed provides sequential and parallel resolution patterns to solve optimization problems through this technique on a wide range of parallel computing platforms. To use the skeletons proposed to solve a particular problem, the skeleton filler has to specify the type of the problem, the type of the solution and the specific characteristics of this paradigm using the C++ programming language, and the user does not need to have any knowledge in parallelizations techniques. This paper describes the sequential and the parallel algorithms for message passing and shared memory platforms, and shows computational results for the 0-1 knapsack problem on an Origin 3000. © Springer-Verlag Berlin Heidelberg 2004.