Parallel adaptive tabu search for large optimization problems

This paper presents a new approach for parallel tabu search based on adaptive par-allelism. Adaptive parallelism demonstrates that massively parallel computing using a hundred of heterogeneous machines is feasible to solve large optimization problems. The parallel tabu search algorithm includes diierent tabu list sizes and new intensiication/diversiication mechanisms. Encouraging results have been obtained in solving the quadratic assignment problem. We have improved the best known solutions for some large real-world problems. 1 Motivation and goals Many interesting combinatorial optimization problems are NP-hard, and then they cannot be solved exactly within a reasonable amount of time. Consequently, heuris-tics must be used to solve real-world problems. Tabu search (TS) is a general purpose heuristic (meta-heuristic) that has been proposed by Glover 1]. TS has achieved widespread successes in solving practical optimization problems in diier-ent domains (resource management, process design, logistics, telecommunications, ...). Promising results of applying TS to a variety of academic optimization problems (traveling salesman, quadratic assignment, time-tabling, job-shop scheduling, ...) are reported in the literature 2]. Solving large problems motivates the development of a parallel implementation of TS. The proliferation of powerful workstations and fast communication networks with constantly decreasing cost/performance ratio have shown the emergence of heterogeneous workstation networks and homogeneous clusters of processors (DEC Alpha farms, IBM SP/2, Cray T3D, CM-5, ...) as platforms for high performance computing. These parallel platforms are generally composed of an important park of machines shared by many users. Many parallel TS algorithms have been proposed in the literature. In general, they don't use advanced programming tools (load balancing, adaptive parallelism, checkpointing, ...) to eeciently use the machines. Most of them are developed for dedicated parallel homogeneous machines. Load analysis of networks of workstations during long periods of time showed that only a few percentage of the available power was used 3]]4]. There is a substantial amount of idle time. In addition, a workstation belongs to an owner who will not tolerate external applications degrading the performance of his machine. Therefore, dynamic adaptive scheduling of parallel applications is essential. Our aim is to develop a parallel adaptive TS strategy, which can beneet greatly from a platform having combined computing resources of massively parallel ma