Task Scheduling in Graphic Processing Units Heterogeneous With Density of The Graph

摘要

An efficient scheduling algorithm is significant to achieve high performance in heterogeneous consisted by the central processing units (CPU) and the graphic processing unit (GPU). However, most of the scheduling algorithms are not suitable to meet the present demands of the heterogeneous. According to the characteristics of the direction acyclic graph, the direction acyclic graph nodes density list scheduling algorithm (DAG-NDLS) was proposed. The algorithm had two important phase, the first phase was task prioritizing, selected the ready task with highest priority, defined by the summation of the node density and the transmission between the node and child nodes, and the second phase was processor selection, selected the processors for the tasks with the earliest finish time minimum. In the end, compared the performances of the proposed algorithm with the heterogeneous earlier finish time algorithm and the high performance task scheduling algorithm, the results of the proposed algorithm were the best in case of the efficiency and utilization.