A Hybrid Evolutionary Heuristic for Job Scheduling on Computational Grids

In this chapter we present a hybrid evolutionary meta-heuristic based on memetic algorithms (MAs) and several local search algorithms. The memetic algorithm is used as the principal heuristic that guides the search and could use any of 16 local search algorithms during the search process. The local search algorithms used in combination with the MA are obtained by fixing either the type of the neighborhood or the type of the move; they include swap/move based search, hill climbing, variable neighborhood search and tabu search. The proposed hybrid meta-heuristic is implemented in C++ using a generic approach based on a skeleton for memetic algorithms. The implementation has been extensively tested in order to identify a set of appropriate values for the MA and local search parameters. We have comparatively studied the combination of MA with different local search algorithms in order to identify the best hybridization. Results are compared with the best known results for the problem in the evolutionary computing literature, namely the benchmark of Braun et al. 2001, which is known to be the most difficult benchmark for static instances of the problem. Our experimental study shows that the MA + TS hybridization outperforms the combinations of MA with other local search algorithms considered in this work and also improves the results of Braun et al. for all considered instances. We also discuss some issues related to the fine tuning and experimenting of meta-heuristics in a dynamic environment.