A Parallel Implementation of a Job Shop Scheduling Heuristic

In the paper we present rst experimental results of a parallel implementation of simulated annealing-based heuristics. The heuristics are developed for the classical job shop scheduling problem that consists of l jobs where each job has to process exactly one task on each of the m machines. We utilize the disjunctive graph representation and the objective is to minimize the length of longest paths, i.e., the overall completion time of tasks. A theoretical parallelization employs O(n 3) processors, where n = l m is the total number of tasks. Since O(n 3) is an extremely large number of processors for real world applications, the heuristics were implemented in a distributed computing environment. The implementation is running on a cluster of 12 processors and has been applied to several benchmark problems. We compare our computational experiments to sequential results and show that stable results equal or close to optimum solutions are calculated by the parallel implementation with a high speed up.