基于EDA多任务流的调度算法研究  

Research on Scheduling Algorithm Based on EDA Multitask Flow

作　　者：王静[1,2,3] 陈岚 张贺[1,3] 王海永 WANG Jing;CHEN Lan;ZHANG He;WANG Haiyong(Institute of Microelectronics of Chinese Academy of Sciences,Beijing 100029,China;University of Chinese Academy of Sciences,Beijing 100049,China;Beijing Key Laboratory of Three-dimensional and Nanometer Integrated Circuit Design Automation Technology,Beijing 100029,China)

机构地区：[1]中国科学院微电子研究所,北京100029 [2]中国科学院大学,北京100049 [3]三维及纳米集成电路设计自动化技术北京市重点实验室,北京100029

出　　处：《计算机工程》2021年第2期146-151,共6页Computer Engineering

基　　金：国家重点研发计划高性能计算专项(2017YFB0203501);北京市科技专项(Z171100001117147)。

摘　　要：针对高性能计算环境下的多个电子设计自动化(EDA)任务流调度问题,提出一种启发式公平调度算法L-Fairness。在准备队列选择待调度任务时,L-Fairness算法结合任务滞后度、完成度和剩余完成时间确定任务优先级,保证多个任务流中任务的公平调度,同时将license数量及其类型作为处理器选择的依据,满足EDA多任务流的调度需求。仿真结果表明,与经典Fairness算法相比,L-Fairness算法的平均资源利用率提高6.7%,不公平度和平均完成时间分别降低46.2%和14.9%,保证了用户服务质量及调度公平性。To address the scheduling problem of multiple Electronic Design Automation(EDA)task flows in high performance computing environment,this paper proposes a heuristic fair scheduling algorithm,L-Fairness. When selecting the tasks to be scheduled in the preparation queue,the algorithm determines the task priority by combining the lag degree,the completion degree,and the remaining completion time to ensure the fair scheduling of tasks in multitask flows. At the same time,the number of licenses and their type are considered during the selection of processor to meet the needs of EDA multitask flow scheduling.Simulation results show that compared with the classic Fairness algorithm,L-Fairness algorithm ensure user Quality of Service(QoS)and scheduling fairness with the average resource utilization rate increased by 6.7%,the unfairness reduced by46.2% and the completion time reduced by14.9%.

关 键 词：电子设计自动化 有向无环图 公平性 资源利用率 license调度 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]