多HPC集群环境中跨群统一作业调度迁移技术  

A united job scheduling and migrating system for high performance computing clusters

作　　者：张毅[1] ZHANG Yi(China Aerodynamics Research and Development Center,Mianyang 621000,China)

机构地区：[1]中国空气动力研究与发展中心,绵阳621000

出　　处：《北京信息科技大学学报（自然科学版）》2021年第2期87-93,共7页Journal of Beijing Information Science and Technology University

摘　　要：为提高计算中心的计算能力和整体资源利用率,分析了多集群环境下各集群的结构组成、节点规模、用户需求、资源管理调度效果及整体性能等,提出了一种针对多套计算集群内有效资源调度的“用户处理器配额抢占策略”,并用Linux编程实现了跨调度软件平台的多集群统一调度迁移机制,介绍了其结构、功能特点和实现方法。模拟测试表明,该调度策略与迁移调度处理系统相结合,可以有效提高计算中心的整体资源利用率和实际操作工作效率。To improve the performance and general resource utilization in the computing center,the structural compositions,nodes scales,needs of users,the effects of resource management and scheduling,and the general performance of multiple clusters are analyzed,a user processor quota preemption policy(UPQPP)is proposed for effective resource scheduling among the high-performance computing(HPC)clusters,and the cross scheduling software platform is implemented with Linux programming,the structure,functional feature,and implementation of which are introduced in this paper.The simulated tests show that the policy combined with the subsystem can effectively improve utilization of computing resources,and improve the operating efficiency of both administrators and users for all computing clusters in the computing center.

关 键 词：高性能计算 多集群环境 计算中心 作业调度 

分 类 号：TP3-0[自动化与计算机技术—计算机科学与技术]