Fast Multi-objective Reschulding of Grid Jobs by Heuristics and Evolution

Scheduling of jobs to a computational grid is a permanent process due to the dynamic nature of the grid and the frequent arrival of new jobs. Thus, a permanent rescheduling of already planned and new jobs must be performed. This paper will continue and extend previous work, which focused on the tuning of our Global Optimising Resource Broker and Allocator GORBA in a static planning environment. A formal definition of the scheduling problem and a classification will be given. New heuristics for rescheduling exploiting the old plan will be introduced and it will be investigated how they contribute to the overall planning process. Furthermore, the maximal possible load, which can be handled within the given time frame of three minutes, will be examined for a grid of growing size of up to 6000 grid jobs and 600 resources.