基于Kubernetes的资源调度策略研究与改进  

Research and improvement of resource scheduling strategy based on Kubernetes

作　　者：于泽川 张娜[1] 包梓群 苏鸿斌 YU Zechuan;ZHANG Na;BAO Ziqun;SU Hongbin(School of Information Science and Technology,Zhejiang Sci-Tech University,Hangzhou 310018,China)

机构地区：[1]浙江理工大学信息学院,杭州310018

出　　处：《智能计算机与应用》2023年第2期1-5,14,共6页Intelligent Computer and Applications

基　　金：国家级大学生创新创业训练计划项目(202010338024);浙江省教育厅一般科研项目(Y202147659);浙江省重点研发计划项目(2020C03094);国家自然科学基金(6207050141)。

摘　　要：针对Kubernetes默认调度策略在多Pod调度时无法考虑多任务调度过程的全局特征,导致无法保证集群整体负载均衡的问题,本文设计了一种优化的静态资源调度策略Ku-PSO,通过改进粒子群算法(PSO),提升集群负载均衡效率。首先,通过建立多Pod调度模型,以集群负载均衡度作为适应函数,并设置约束条件保证集群正常运行;其次,通过改进粒子群算法的惯性因子、个体学习因子和社会学习因子实现权值优化,使得粒子群在前期寻优过程中具有优秀的全局搜索能力,后期寻优过程中能够迅速收敛,应用于集群资源调度能够快速找出资源的最优分配方案。实验表明,使用Ku-PSO算法进行Kubernetes资源调度较默认调度策略集群均衡度显著提升,较PSO算法可以有效减少部署时间,实现更优的均衡调度。Aiming at the problem that the Kubernetes scheduling strategy cannot consider the global characteristics of the multi-task scheduling process when scheduling multiple Pods,resulting in the inability to ensure the overall load balance of the cluster,this paper designs an optimized static resource scheduling strategy Ku-PSO.It incorporates the particle swarm algorithm(PSO)to improve cluster load balancing efficiency.First,by establishing a multi-Pod scheduling model,the cluster load balance degree is used as the adaptation function,and constraints are set to ensure the normal operation of the cluster.Then,by improving the inertia factor,individual learning factor and social learning factor of the particle swarm algorithm with the weight optimization,the particle swarm has excellent global search ability the in the early optimization process.In later optimization process,the local optimization ability is improved,and it can converge quickly.When applied to cluster resource scheduling,it can quickly find the optimal allocation scheme of resources.Experiments show that the use of the Ku-PSO algorithm for Kubernetes resource scheduling significantly improves the cluster balance of the default scheduling strategy,and can effectively reduce the deployment time compared with the PSO algorithm and achieve better balanced scheduling.

关 键 词：Kubernetes 资源调度 粒子群算法 权值优化 均衡调度 

分 类 号：TP311.5[自动化与计算机技术—计算机软件与理论]