Fair scheduling in grid VOs with anticipation heuristic

Abstract

In this work, a job-flow scheduling approach for Grid virtual organizations (VOs) is proposed and studied. Users’ and resource providers’ preferences, VOs internal policies along with local private utilization impose specific requirements for scheduling according to different, usually contradictive, criteria. We study the problem of a fair job batch scheduling with a relatively limited resources supply. With increasing resources utilization level the available resources set and corresponding decision space are reduced. The main problem is a scarce set of job execution alternatives which eliminates scheduling optimization. In order to improve overall scheduling efficiency we propose a heuristic anticipation approach. It generates a reference, most likely infeasible, scheduling solution. A special replication procedure performs a feasible solution with a minimum distance to a reference alternative under given metrics.