Efficient scheduling of DAG tasks on multi-core processor based parallel systems

Directed Acyclic Graph (DAG) task models have been widely used in multi-core processor based parallel system. The efficiency of it is closely related to the performance of multi-core processors. However, most of the researches ignore the allocation of the non-critical predecessors, so that they still have some disadvantages. The number of processors and the schedule length need to be further reduced. In order to solve this problem, we propose a new scheduling algorithm, which combined the cluster-based method and the interval insertion strategies. The earliest start time will be obtained according to the cluster-based method. Strategies of task-duplication and interval insertion will also be given to satisfy the earliest start time. In this paper, the non-critical predecessors can be assigned to the same processor with the key predecessor or the same processor with other non-critical predecessors. So the schedule length and the number of processors can be both reduced by proposed method, especially the situation of high communication cost. More efficient task scheduling can be achieved.