Local Search for Deadline Driven Grid Scheduling

This work describes application of local search based algorithms for job scheduling in the Grid environment where dynamic changes occur. The primary intent is to consider problems with the typical quality of service constraint taking into account and minimizing the number of late jobs. To achieve this goal a special instance of the Tabu search algorithm applicable to dynamic problems is proposed. Also a new dispatching rule generating the initial solution is introduced. Comparison with typical queue-based policies such as First Come First Served, Earliest Deadline First or Easy Backfilling is provided. Experimental results shows that local search based algorithms is a promising technique with better performance than queue-based algorithms while still fast enough to provide solutions in a reasonable time.