高性能计算集群实验室的构建与优化研究  

Research on the Construction and Optimization of High-Performance Computing Cluster Laboratory

作　　者：孙海松 SUN Haisong(Department of Information Engineering,Nanjing Polytechnic Institute,Nanjing Jiangsu 210048,China)

机构地区：[1]南京科技职业学院信息工程学院,江苏南京210048

出　　处：《信息与电脑》2022年第10期8-10,共3页Information & Computer

摘　　要：本文构建了面向计算密集型任务的高性能计算集群实验室(High Performance Computing Cluster Lab,HPCCL),介绍了该集群的主要硬件和软件配置。为了提高集群实验室的管理效率,结合深度强化学习技术,提出了计算作业管理优化方法,即JoSHH (Job Scheduling for HPCCL)。基于所构建的集群实验室,进行实验评估该方法的性能。实验结果表明,与现有方法相比,该方法减少了计算任务的运行时间。This paper constructs a High Performance Computing Cluster Lab(HPCCL) for computing intensive tasks,and introduces the main hardware and software configurations of the cluster.In order to improve the management efficiency of the cluster laboratory,combined with the deep reinforcement learning technology,a computational job management optimization method,namely JoSHH(Job Scheduling For HPCCL),is proposed.Based on the cluster laboratory,the performance of this method is evaluated by experiments.Experimental results show that compared with existing methods,this method reduces the running time of computing tasks.

关 键 词：高性能计算集群 深度学习 作业管理优化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]