开源云上的Kubernetes弹性调度
|
摘要
针对私有云资源弹性调度问题,将Kubernetes结合已有Openstack云平台,提出一种基于容器的弹性调度策略。一方面,因为Openstack虚拟机启动时间较长,给调度带来额外时间开销,所以利用容器拉起时耗远小于虚拟机的特性,用Docker容器取代了Openstack默认的虚拟机;一方面优化了Kubernetes调度算法,建立了一个提高集群资源利用率的优化模型,通过对云平台各个服务器节点四种类型资源的监控和应用队列预设模板匹配,选择调度资源利用率最高的服务器。整个调度过程包括容器应用的初次调度和在线迁移算法。实验结果表明,相比原有Kubernetes调度算法和一些其他的调度策略,该调度策略对数据中心资源进行了更细粒度的划分,在保证服务器性能的同时,实现了云平台资源弹性调度,集群资源利用率也得到了提高,同时降低了数据中心能耗。
Aiming at the problem of flexible scheduling of private cloud resources,Kubernetes combined with the existing Openstack cloud platform,we propose a flexible scheduling strategy based on container.On the one hand,because Openstack virtual machine takes a long time to start up and brings extra time cost to scheduling,Docker container is used to replace the default virtual machine of Openstack by taking advantage of the feature that container takes much less time to pull up than virtual machine.On the one hand,the Kubernetes scheduling algorithm is optimized,and an optimization model to improve the utilization rate of cluster resources is established.By monitoring four types of resources of each server node of the cloud platform and applying queue preset template matching,the server with the highest utilization rate of scheduling resources is selected.The whole scheduling process includes the initial scheduling of container applications and the online migration algorithm.Experiment shows that compared with the original Kubernetes scheduling algorithm and some other scheduling strategies,this scheduling strategy divides the data center resources into finer granularity.While ensuring the server performance,it realizes the flexible scheduling of cloud platform resources,improves the utilization rate of cluster resources and reduces the energy consumption of the data center.
Aiming at the problem of flexible scheduling of private cloud resources,Kubernetes combined with the existing Openstack cloud platform,we propose a flexible scheduling strategy based on container.On the one hand,because Openstack virtual machine takes a long time to start up and brings extra time cost to scheduling,Docker container is used to replace the default virtual machine of Openstack by taking advantage of the feature that container takes much less time to pull up than virtual machine.On the one hand,the Kubernetes scheduling algorithm is optimized,and an optimization model to improve the utilization rate of cluster resources is established.By monitoring four types of resources of each server node of the cloud platform and applying queue preset template matching,the server with the highest utilization rate of scheduling resources is selected.The whole scheduling process includes the initial scheduling of container applications and the online migration algorithm.Experiment shows that compared with the original Kubernetes scheduling algorithm and some other scheduling strategies,this scheduling strategy divides the data center resources into finer granularity.While ensuring the server performance,it realizes the flexible scheduling of cloud platform resources,improves the utilization rate of cluster resources and reduces the energy consumption of the data center.
引文
[1]刘鹏.云计算[M].第3版.北京:电子工业出版社,2011.
[2]TURNBULL J.The Docker book:containerization is the new virtualization[M].[s.l.]:[s.n.],2014:4-7.
[3]CASALICCHIO E.Autonomic orchestration of containers:problem definition and research challenges[C]//EAI international conference on performance evaluation methodologies and tools.[s.l.]:[s.n.],2017.
[4]陈耿.开源容器云openshift构建基于kubernetes的企业应用云平台[M].北京:机械工业出版社,2017.
[5]NETTO H V,LUNG L C,CORREIA M,et al.State machine replication in containers managed by Kubernetes[J].Journal of Systems Architecture,2017,73:53-59.
[6]龚正.Kubernetes权威指南:从Docker到Kubernetes实践全接触[M].北京:电子工业出版社,2015.
[7]KOMINOS C G,SEYVET N,VANDIKAS K.Bare-metal,virtual machines and containers in OpenStack[C]//20th conference on innovations in clouds,internet and networks.Paris,France:IEEE,2017:36-43.
[8]FELTER W,FERREIRA A,RAJAMONY R,et al.An updated performance comparison of virtual machines and Linux containers[C]//IEEE international symposium on performance analysis of systems and software.Philadelphia,PA,USA:IEEE,2014:438-453.
[9]CHEN C,REN S,HAU E.A Kubernetes-based monitoring platform for dynamic cloud resource provisioning[C]//IEEE global communications conference.[s.l.]:IEEE,2017:1-6.
[10]董西成.Hadoop技术内幕:深入解析YARN架构设计与实现原理[M].北京:机械工业出版社,2014.
[11]唐瑞.基于Kubernetes的容器云平台资源调度策略研究[D].成都:电子科技大学,2014.
[12]杨鹏飞.基于Kubernetes的资源动态调度的研究与实现[D].杭州:浙江大学,2017.
[13]彭丽苹,吕晓丹,蒋朝惠,等.基于Docker的云资源弹性调度策略[J].计算机应用,2018,38(2):557-562.
[14]董西成.Hadoop技术内幕:深入解析MapReduce架构设计与实现原理[M].北京:机械工业出版社,2013.
[15]董春涛,李文婷,沈晴霓,等.Hadoop YARN大数据计算框架及其资源调度机制研究[J].信息通信技术,2015,9(1):77-84.
[2]TURNBULL J.The Docker book:containerization is the new virtualization[M].[s.l.]:[s.n.],2014:4-7.
[3]CASALICCHIO E.Autonomic orchestration of containers:problem definition and research challenges[C]//EAI international conference on performance evaluation methodologies and tools.[s.l.]:[s.n.],2017.
[4]陈耿.开源容器云openshift构建基于kubernetes的企业应用云平台[M].北京:机械工业出版社,2017.
[5]NETTO H V,LUNG L C,CORREIA M,et al.State machine replication in containers managed by Kubernetes[J].Journal of Systems Architecture,2017,73:53-59.
[6]龚正.Kubernetes权威指南:从Docker到Kubernetes实践全接触[M].北京:电子工业出版社,2015.
[7]KOMINOS C G,SEYVET N,VANDIKAS K.Bare-metal,virtual machines and containers in OpenStack[C]//20th conference on innovations in clouds,internet and networks.Paris,France:IEEE,2017:36-43.
[8]FELTER W,FERREIRA A,RAJAMONY R,et al.An updated performance comparison of virtual machines and Linux containers[C]//IEEE international symposium on performance analysis of systems and software.Philadelphia,PA,USA:IEEE,2014:438-453.
[9]CHEN C,REN S,HAU E.A Kubernetes-based monitoring platform for dynamic cloud resource provisioning[C]//IEEE global communications conference.[s.l.]:IEEE,2017:1-6.
[10]董西成.Hadoop技术内幕:深入解析YARN架构设计与实现原理[M].北京:机械工业出版社,2014.
[11]唐瑞.基于Kubernetes的容器云平台资源调度策略研究[D].成都:电子科技大学,2014.
[12]杨鹏飞.基于Kubernetes的资源动态调度的研究与实现[D].杭州:浙江大学,2017.
[13]彭丽苹,吕晓丹,蒋朝惠,等.基于Docker的云资源弹性调度策略[J].计算机应用,2018,38(2):557-562.
[14]董西成.Hadoop技术内幕:深入解析MapReduce架构设计与实现原理[M].北京:机械工业出版社,2013.
[15]董春涛,李文婷,沈晴霓,等.Hadoop YARN大数据计算框架及其资源调度机制研究[J].信息通信技术,2015,9(1):77-84.