基于G-EDF的DAG并行任务多核响应时间分析  

Response Time Analysis of Multiprocessor Systems for DAG Parallel Tasks Based on G-EDF

作　　者：韩美灵 邓庆绪 张天宇 林宇晗 HAN Mei-ling;DENG Qing-xu;ZHANG Tian-yu;LIN Yu-han(School of Computer Science & Engineering,Northeastern University,Shenyang 110169,China)

机构地区：[1]东北大学计算机科学与工程学院,辽宁沈阳110169

出　　处：《东北大学学报（自然科学版）》2019年第3期315-320,共6页Journal of Northeastern University(Natural Science)

基　　金：国家自然科学基金资助项目(61472072;61528202);辽宁重大装备制造协同创新中心资助项目

摘　　要：由于未考虑DAG(directed acyclic graph)任务的自身结构,基于G-EDF(global earliest deadlinefirst)的DAG并行任务模型的可调度性分析存在很大的悲观性,因此本文针对DAG任务集在多处理器系统中采用G-EDF调度策略下的响应时间分析进行了研究.首先针对carry-in任务实例执行的情况提出更加精确的carry-in工作量估算方法.基于该carry-in工作量估算方法提出一种基于完成时间的问题窗口工作量估算方法.最后,结合上述两个改进策略提出了基于G-EDF的DAG任务响应时间分析方法.仿真实验表明,所提出的方法较目前已知的调度策略方法可调度性至少提高15%,最高可达25%.Since the self-structure of the DAG(directed acyclic graph)task is not considered,the schedulability analysis of the DAG parallel task model based on G-EDF(global earliest deadline first)is very pessimistic.The response time analysis of the DAG task set under the G-EDF scheduling strategy in multiprocessor systems was studied in this paper.First,a more accurate carry-in workload estimation method was proposed for the execution of the carry-in task instance.Then a method for estimating the problem window workload of completion time was put forward based on the carry-in workload estimation method.Based on the two proposed methods,this paper proposed a response time analyzing method to derive a response time upper bound of each task.The experiments show that the proposed method outperforms the state-of-the-art method by at least 15% and at most 25%.

关 键 词：嵌入式实时系统 多核处理器 并行任务模型 全局调度 响应时间分析 

分 类 号：TP316.2[自动化与计算机技术—计算机软件与理论]