A GA+TS Hybrid Algorithm for Independent Batch Scheduling in Computational Grids

The hybridization of heuristics methods has been proposed in the meta-heuristics research community as a way to overcome limitations of single meta-heuristics. Although a priori there is no guarantee that the resulting hybrid heuristic would outperform stand alone heuristics, in general better results can be found for classes of instances of the optimization problem under study. One promising family of hybridization algorithms is that of population based heuristics with local search heuristics. In this paper we present a high level hybridization of Genetic Algorithms (GAs) and Tabu Search (TS), denoted GA+TS algorithm, for scheduling in computational grids. In the job scheduling problem, the objective is to efficiently compute a planning of incoming jobs to available machines in the Grid system so that the system performance is optimized. Our GA+TS high level hybrid algorithm runs first the GA flow, until a stopping condition is met, and then passes the best output solution in input as starting solution to TS algorithm for further improvement, the flow of TS is then executed. The objective is thus to sequence the calls of GA and TS in a chain that yields to efficient Grid schedulers that would eventually perform better than the same meta-heuristics used as simple schedulers without any additional support and improvement mechanism of the local solutions. We evaluated the proposed hybrid algorithm using different grid scenarios generated by a grid simulator. The computational results showed that the hybrid GA+TS algorithm outperforms both the GA and TS for some classes of instances of the Grid scheduling problem.