KubeTea:面向容器云环境的轻量级多维度微服务应用调度框架  

KubeTea:lightweight multi-dimensional microservice application schedulingframework for container cloud environments

作　　者：李宗霖 何俊江 李汶珊 吕虓 兰小龙 李涛 Li Zonglin;He Junjiang;Li Wenshan;Lyu Xiao;Lan Xiaolong;Li Tao(School of Cyber Science&Engineering,Sichuan University,Chengdu 610065,China;School of Cybersecurity,Chengdu University of Information Technology,Chengdu 610225,China)

机构地区：[1]四川大学网络空间安全学院,成都610065 [2]成都信息工程大学网络空间安全学院,成都610225

出　　处：《计算机应用研究》2024年第9期2794-2799,共6页Application Research of Computers

基　　金：国家重点研发计划资助项目(2020YFB1805400);国家自然科学基金资助项目(62032002,62101358);四川省自然科学青年基金资助项目(2023NSFSC1395);四川大学专职博士后研发基金资助项目(2023SCU12127)。

摘　　要：容器云中,应用和资源调度始终是集群管理的重点。如何在提高资源利用率的同时保证应用服务质量是目前行业积极探索的问题之一。针对该问题,提出一个面向容器云环境的轻量级多维度微服务应用调度框架。该框架设计了非侵入式的网络调用观测方法,并基于观测和监控数据定义亲和性、热点值(HV)和热路径(HP)来指导调度决策。为平衡资源利用率和应用QoS,该框架在调度的垂直方向上提出弹性余量控制机制(elastic slack controller,ESC),在水平方向上考虑了微服务亲和性;并设计了扩散导向的自动伸缩策略(diffusion-oriented autoscaling strategy,DOAS)以缓解应用出现的QoS下降。实验表明,该框架与主流的Kubernetes原生调度工具相比,在集群资源利用率方面提高21%,同时能降低23%的应用端到端时延,实现资源利用率和应用QoS的平衡。In container clouds,application and resource scheduling has always been a key focus of cluster management.One of the active explorations in the industry is how to improve resource utilization while ensuring application service quality.This paper proposed a lightweight microservice application scheduling framework for container cloud environments.This framework designed a non-intrusive network call observation method and defined affinity,hot value(HV),and hot path(HP)based on observation and monitoring data to guide scheduling decisions.To balance resource utilization and application of QoS,the framework proposed the ESC mechanism in the vertical direction of scheduling and considered microservice affinity in the horizontal direction.Furthermore,it designed the DOAS to alleviate the QoS degradation of applications.Experimental results show that compared with the mainstream Kubernetes native scheduling tool,this framework improves resource utilization by 21%while reducing end-to-end application latency by 23%,achieving a balance between resource utilization and QoS application.

关 键 词：资源调度 容器 微服务 可观测性 云原生 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]