基于主机性能度量的异构集群作业调度算法  

Job Scheduling Algorithm Based on Host Performance Measurement in Heterogeneous Clusters

作　　者：王超 陈刚[1,2] 黄刚 王彪[1,2] WANG Chao;CHEN Gang;HUANG Gang;WANG Biao(State Key Laboratory ofAstronautic Dynamics,Xi'an 710043,China;Xi'an Satellite Control Center,Xi'an 710043,China)

机构地区：[1]宇航动力学国家重点实验室,陕西西安710043 [2]西安卫星测控中心,陕西西安710043

出　　处：《计算机与网络》2021年第8期67-73,共7页Computer & Network

摘　　要：传统经典作业度算法在集群应用中实现简单、执行效率高,但在异构集群环境下由于缺乏在线节点运行状态动态反馈能力和负载均衡能力,降低了计算资源利用率和系统吞吐率。为解决上述问题,设计了一种在异构集群环境下基于主机性能度量的作业负载均衡调度算法,该算法通过收集集群中在线节点的状态信息和作业响应时间遴选出可信节点集合,计算出各可信节点的HPM值,利用负载均衡运算规则生成候选的作业分配节点集合,最终按照预先设计的优先原则把不同作业分配至各计算节点,并更新各节点运行状态。实验结果表明,在异构集群环境下调度同类型作业时,该算法在总完成时间和负载均衡性能等指标上均优于传统经典算法。The traditional job scheduling algorithms are easy to implement and have high efficiency in cluster applications.However,in heterogeneous cluster environment,due to the lack of dynamic feedback and load balancing capabilities of online nodes running status,the utilization of computing resources and system throughput are reduced.To solve this problem,a job host performance measurement load balancing(HPMLB)algorithm based on host performance measurement(HPM)in heterogeneous clusters is designed.The algorithm collects the status information and the job response time of the online node to select the trusted node,then calculates the HPM value of each trusted node,generating a candidate set through the load balancing operation rule,finally assigns the task according to a certain priority principle,and updates the status of each node.The experimental results show that the HPMLB algorithm outperforms the traditional algorithm in terms of total completion time and load balancing performance when scheduling the same type of jobs in heterogeneous cluster environment.

关 键 词：异构集群 主机性能度量 负载均衡 作业调度 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]