超大规模计算平台-感知混合容器集群的高性能计算作业调度  

High performance computing job scheduling for Ultra Large Scale Computing Platform perception hybrid container clusters

作　　者：董爱强[1] 胡学勇[1] 于兴江 刘旭 戴发玉 DONG Aiqiang;HU Xueyong;YU Xingjiang;LIU Xu;DAI Fayu(Beijing China-Power Information Technology Co.,Ltd.,Beijing 100192,China;State Grid Info-Telecom Great Power Science And Technology Co.,Ltd.,Fuzhou 350003,China)

机构地区：[1]北京中电普华信息技术有限公司,北京海淀100192 [2]国网信通亿力科技有限责任公司,福建福州350003

出　　处：《自动化与仪器仪表》2024年第10期60-64,共5页Automation & Instrumentation

基　　金：国网信息通信产业集团有限公司科技创新项目,电力数字化基础平台及关键组件研发(重点专项)《面向能源行业的自主可控“思极云”平台关键技术研究与应用》(5468B1230001)。

摘　　要：针对高性能计算作业在混合容器集群中调度效率低下的问题,研究提出ULSCP-Perception调度策略。该策略通过优化资源分配,以解决传统Kubernetes集群作业调度对资源利用率不高的问题,并提升作业执行效率。基于混合容器集群的实际作业调度,结果表明ULSCP-Perception策略不仅可以将作业的平均执行时间从1549.8 s降至578.37 s,还能够降低CPU占用率,平均降幅达32.37%。对于云计算与容器技术融合下的HPC作业调度,研究具有指导云基础设施中资源调度策略实施的实际价值。它不仅有助于突破传统Kubernetes集群在资源调度上的局限,还为HPC作业调度带来了新的视角,促进了作业调度策略向更加智能化和高效化的发展。此外,这种策略的实施对于降低运营成本、提高服务质量、增强用户满意度都有直接的推动作用。Aiming at the problem of low scheduling efficiency of high-performance computing jobs in hybrid container clusters,the ULSCP Perception scheduling strategy is proposed.This strategy optimizes resource allocation to solve the problem of low resource utilization in traditional Kubernetes cluster job scheduling and improves job execution efficiency.Based on the actual job scheduling of mixed container clusters,the results show that the ULSCP Perception strategy can not only reduce the average execution time of jobs from 1549.8 seconds to 578.37 seconds,but also reduce CPU usage,with an average reduction of 32.37%.The research on HPC job scheduling under the integration of cloud computing and container technology has practical value in guiding the implementation of resource scheduling strategies in cloud infrastructure.It not only helps to break through the limitations of traditional Kubernetes clusters in resource scheduling,but also brings a new perspective to HPC job scheduling,promoting the development of job scheduling strategies towards greater intelligence and efficiency.In addition,the implementation of this strategy has a direct driving effect on reducing operating costs,improving service quality,and enhancing user satisfaction.

关 键 词：ULSCP-Perception 混合容器集群 HPC 作业调度 感知决策 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]