Performance Improvement of Genetic Algorithms by Adaptive Grid Workflows

In this paper we present improvement of the performance of Grid Direct Acyclic Graph (DAG) workflow genetic algorithm by harnessing the power of High Level Petri-Nets workflow model. Genetic Algorithms are very powerful optimization technique that is easily parallelized using different approaches which makes it ideal for the Grid. The High Level Petri-Net workflow model greatly outperforms currently available DAG workflow model available in gLite Grid middleware. Using the flexibility of the High Level Petri-Net workflows we have designed an adaptive workflow that overcomes the heterogeneity and unpredictability of the Grid infrastructure, giving users better and more stable execution times than formerly used DAG workflows. The experimental results obtained by Genetic Algorithm optimization of performance of the Data Warehouse design have shown advantages of the new approach by shortening the optimization time up to 50% for the same CPU time utilization. Another advantage is the increased stability of the time variance of the estimated execution time to approximately 30 minutes for runs on different Grid loads.