A Hybrid Task Scheduler for DAG Applications on a Cluster of Processors

With the advent of large scale computing environment, there is a need for scheduling methods which allow multiple DAG-structured applications to share the cluster resources with the objectives of increasing the throughput and maximizing the resource utilization. The number of processors allotted to each application is decided during runtime, depending on the number and nature of applications being scheduled together. The static schedule generated is fixed for a specified number of processors and also the quality of schedule is handicapped by the estimated execution time of tasks in the application. Since the number of processors allotted for each DAG is not known a priori in a multiple DAG scheduling environment, instead of a static schedule, a dynamic scheduler which relies on actual execution time of tasks is better. But its performance is limited as the only information it has is the number of tasks scheduled on each processor. The knowledge of an application in terms of its structure, estimated execution time of tasks and relative importance of tasks, can be used by a dynamic scheduler for better scheduling of tasks resulting in decreased application execution time. With the use of a priori application information, different scheduling policies for a hybrid scheduler are proposed and their performance is compared with the scheduler found in literature. The results obtained demonstrate the suitability of the proposed hybrid scheduler in multiple DAGs scheduling environment.