基于混合集群的高性能计算作业感知与调度研究  

Research on High Performance Computing Job Awareness and Scheduling Based on Hybrid Cluster

作　　者：粟海斌 刘斌 张冠澜 车宇烽 詹柱 SU Haibin;LIU Bin;ZHANG Guanlan;CHE Yufeng;ZHAN Zhu(Fangxin Technology Co.,Ltd.,Changsha Hunan 410000)

机构地区：[1]方心科技股份有限公司,湖南长沙410000

出　　处：《中国科技纵横》2025年第3期56-58,共3页China Science & Technology Overview

摘　　要：针对高性能计算在资源有限、成本高昂等方面面临的挑战,本文设计并实现了一种面向混合集群的作业感知与调度框架。该框架集成了感知策略、调度策略、协调模块和环境监控模块,以解决高性能计算应用日益增多带来的问题。通过该框架,本文对资源负载预测模型和高性能计算作业调度策略在混合集群数据中心的应用进行了验证,旨在降低集群能耗和减少经济成本。研究结果表明,混合集群环境下的调度策略能够有效提高资源利用率和作业执行效率,降低成本和能耗,为研究人员和HPC用户提供更经济、高效的计算资源使用方式,推动HPC技术在社会各个领域的应用和发展。Aiming at the challenges of high performance computing such as limited resources and high cost,this study designs and implements a job awareness and scheduling framework for hybrid clusters.The framework integrates sensing strategy,scheduling strategy,coordination module and environment monitoring module to solve the problems caused by the increasing number of HPC applications.Through this framework,this study verifies the application effect of the proposed resource load prediction model and HPC job scheduling strategy in the hybrid cluster data center,aiming at reducing cluster energy consumption and reducing economic costs.The results show that the scheduling strategy in the hybrid cluster environment can effectively improve resource utilization and job execution efficiency,while reducing costs and energy consumption.It provides a more economical and efficient way to use computing resources for researchers and HPC users,and helps to promote the application and development of HPC technology in various fields of society.

关 键 词：高性能计算 作业感知 调度策略 资源监控 能耗降低 

分 类 号：TP38[自动化与计算机技术—计算机系统结构]