The ParMetaOpt Experience: Performance of Parallel Metaheuristics on Scheduling Optimization PLAMENKA BOROVSKA, GEORGE YANCHEV Computer Systems Department Technical University of Sofia 8 Kliment Ohridski Boul.,1756 Sofia BULGARIA pborovska@tu-sofia.bg http://csconf.org/cs/leader_eng.htm jyanchev@dir.bg Abstract: - Parallel metaheuristics provides innovative and powerful alternative for combinatorial optimization providing the opportunity to find out near-optimal solutions in reasonable time. The goal of this paper is to reveal the experience of utilizing the experimental parallel metaheuristics framework ParMetaOpt, developed at Computer Systems Dept., Technical University of Sofia. Parallel metaheuristics algorithm have been designed and implemented based on population based methods (evolutionary computation, artificial bee colony and ant colony optimization) and trajectory based methods (GRASP, Tabu search and simulated annealing) for the case studies of the timetabling and the job shop scheduling problems. Parallel performance evaluation and analysis have been presented on the basis of hybrid (MPI+OpenMP) parallel program implementations on compact heterogeneous computer cluster. Key-Words: Metaheuristics, Combinatorial Optimization, Scheduling Optimization, Parallel Algorithms and Programming, Parallel Performance 1 Introduction Modern metaheuristics [1,4] has emerged as an innovative and powerful alternative that offers a wide spectrum of algorithmic frameworks providing the opportunity for IT professionals to solve combinatorial (NP-complete) problems in reasonable time and finding out solutions of good quality. Parallelizing metaheuristics [2] is a grand challenge due to the fact that the fundamental concepts of metaheuristics – diversification and intensification – predetermine the possibility to deploy parallel computing strategies to obtain better performance and to improve the quality of solution as well. ParMetaOpt is an experimental parallel metaheuristics framework [3], developed at Computer Systems Dept., Technical University of Sofia, that is intended to provide flexible components for implementing parallel metaheuristics algorithms for combinatorial optimization based on object oriented programming techniques. The computing platform comprises heterogeneous compact computer cluster comprising dual-core Opteron-based servers and quad- core Xeon-based servers, the operating system is Scientific Linux, the programming language is C++ in combination with the standard application programming interfaces MPI and OpenMP. The experimental framework is used by researchers, PhD students and MSc students. The goal of this paper is to share our experience in solving scheduling optimization problems such as the timetabling and the job shop scheduling problems on the ParMetaOpt parallel metaheuristics framework in respect to parallel performance. 2 Scheduling Optimization Problem Instances Scheduling optimization problems are hard and time consuming to solve. Scheduling problems deal with the temporary planning of activities with limited resources. Depending on the specific problem the activities may be jobs, tasks, lectures, trains, etc. The corresponding resources may comprise machines, processors, teachers, etc. The case studies of scheduling optimization under investigation are the course timetabling problem and the job shop scheduling problems 2.1 The course timetabling problem The course timetabling problems require the assignment of a set of events such as lectures, labs, etc. to a limited number of time slots satisfying predefined constraints. The solution suggests a sequence of meetings between teachers and students in a prefixed period of time (typically a week) in the available classrooms and auditoria, satisfying a set of constraints of various types Proceedings of the 9th WSEAS International Conference on APPLIED INFORMATICS AND COMMUNICATIONS (AIC '09) ISSN: 1790-5109 475 ISBN: 978-960-474-107-6