摘要
在单芯片多核系统中,No C已成为主流片上通信架构。有效的任务调度是挖掘计算并行性的重要方法。在经典静态列表调度基础上,针对HEFT算法中节点排序会得出较多的优先级相同节点的问题,提出一种节点二次排序的调度方法。在边的调度上应用了ALAP原则,改进算法有效提高了调度效果。实验表明,新方法对bl、blcomp、blio等节点优先权算法得出的任务列表均有良好的调度效果,适应性较好;对于2D Mesh同构No C架构,改进算法对三种节点优先权算法有1.15倍的平均加速比,最大可有1.27倍加速比。
No C becames the main communication architecture of MPSOC system. As its hardware being mature,the scheduling technology for No C becomes the key factors to restrict the performance of the multi-core. It has been becoming the hot point to research the more appropriate scheduling algorithm to make full parallelism of multi-core. Considering the efficiency of classical HEFT algorithm decreased by many nodes with the same priority,this paper presented a novel scheduling algorithm to gain the better scheduling result by ranking those same priority nodes again and application of the rule of ALPA in edges scheduling. The proposed algorithm has good applicability to schedule lists derived from bl、blcomp、blio and a variety of node priority. The experiment result of 2D Mesh No C shows that the average and the maximum speedup compared with the HEFT algorithm is 1. 15 and 1. 27.
出处
《计算机应用研究》
CSCD
北大核心
2016年第4期1054-1057,共4页
Application Research of Computers
基金
国家自然科学基金面上项目(61179036)
关键词
任务调度
静态调度
片上网络
有向无环图
task scheduling
static scheduling
network on chip（NoC）
directed acyclic graph（DAG）
