NUMA感知的云平台负载调度系统  

NUMA-aware cloud platform workload scheduling system

作　　者：娄杰 段宏键 曹华伟 叶笑春[1,2] LOU Jie;DUAN Hongjian;CAO Huawei;YE Xiaochun(State Key Lab of Processors,Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190;School of Computer Science and Technology,University of Chinese Academy of Sciences,Beijing 100049)

机构地区：[1]处理器芯片全国重点实验室(中国科学院计算技术研究所),北京100190 [2]中国科学院大学计算机科学与技术学院,北京100049

出　　处：《高技术通讯》2025年第1期20-36,共17页Chinese High Technology Letters

基　　金：国家重点研发计划(2022YFB4501404);北京市自然科学基金(4232036)资助项目。

摘　　要：随着互联网的高速发展,云计算逐渐走向了云原生时代。在云原生领域中,对容器进行调度与编排的标准系统是Kubernetes。Kubernetes有着开源、可扩展、部署难度低等诸多优点,然而,随着容器化应用的多样化和底层资源的多元化,Kubernetes在以非统一存储访问(non-uniform memory access,NUMA)资源为代表的细粒度资源调度方面仍然存在不足,集群中计算资源利用率低、使用不均衡、系统关键资源争用等情况常常发生。本文以Kubernetes系统为基础,探究以NUMA为代表的细粒度资源的优化调度机制,具体研究点如下:(1)建立缓存管理器,对集群中基于容器的典型应用进行性能的建模与特征分析;(2)设计NUMA管理器,实现细粒度资源划分;(3)优化面向细粒度资源调度的算法,细粒度分配NUMA资源。通过NUMA感知的调度优化,本文所提方案提高了系统的关键资源利用率,提升了应用的运行速度,减少了集群中资源的争用以及资源使用上不均衡的现象。With the rapid growth of the Internet,cloud computing is transitioning into the cloud-native era.In the cloudnative landscape,Kubernetes serves as a standard system for container scheduling and orchestration.Kubernetes offers numerous advantages such as open-source,scalability,and easiness of deployment.However,as containerized applications diversify and underlying resources become more heterogeneous,Kubernetes still faces challenges in fine-grained resource scheduling,particularly regarding non-uniform memory access(NUMA)resources.Uneven utilization of computing resources and contention for critical system resources are common in clusters.Based on the Kubernetes system,this article explores an optimized scheduling mechanism for fine-grained resources,represented by NUMA.The specific research areas are as follows.(1)Establishing a cache manager to model and analyze the performance of typical container-based applications in the cluster.(2)Designing a NUMA manager to implement fine-grained resource partitioning.(3)Optimizing algorithms for fine-grained resource scheduling and allocating fine-grained NUMA resources.Through NUMA-aware scheduling optimization,the proposed scheme enhances the utilization of critical resources,improves application performance,and reduces resource contention as well as imbalances in the cluster.

关 键 词：云计算 容器云平台 负载调度 非统一存储访问 资源划分 

分 类 号：TP393.09[自动化与计算机技术—计算机应用技术]