RainbowD:一种异构云环境下高效的Docker镜像分发系统  

RainbowD:A Heterogeneous Cloud-Oriented Efficient Docker Image Distribution System

作　　者：张浩[1,2] 孙毓忠 肖立[1,2] 唐勇 胡满满 杜沁园 蔡志彬 冯百明 ZHANG Hao;SUN Yu-Zhong;XIAO Li;TANG Yong;HU Man-Man;DU Qin-Yuan;CAI Zhi-Bin;FENG Bai-Ming(Institute of Computing Technology Chinese Academy of Sciences,Beijing 100190;University of Chinese Academy of Sciences,Beijing 100049;College of Computer Science&Engineering,Northwest Normal University,Lanzhou 730079;School of Software Engineering,University of Science and Technology of China,Hefei 230000;School of Computer and Information Engineering,Beijing Technology and Business University,Beijing 100048)

机构地区：[1]中国科学院计算技术研究所,北京100190 [2]中国科学院大学,北京100049 [3]西北师范大学计算机科学与工程学院,兰州730070 [4]中国科学技术大学软件学院,合肥230000 [5]北京工商大学计算机与信息工程学院,北京100048

出　　处：《计算机学报》2020年第11期2067-2083,共17页Chinese Journal of Computers

基　　金：“面向云计算的网络化操作系统”(2016YFB1000505);国家自然科学基金委员会(NSFC)-广东省人民政府联合基金超级计算科学应用研究专项计划(第二期)(U1611261)资助.

摘　　要：近年来,互联网技术飞速发展,互联网用户的数量呈指数级增长;随之而来的是用户需求的服务数量越来越大、服务种类也越来越多.不仅如此,虚拟化技术也在推陈出新,Docker以其轻量性、可移植性好等优良特性备受开发者青睐,Docker在虚拟化技术中扮演越来越重要的角色;为了提升服务的部署效率,越来越多的人选择使用Docker在云环境中部署服务.在集群中部署服务时需要将具有对应服务的镜像分发到节点上,如何高效地分发Docker镜像成了研究的热点问题.然而,在当代互联网发展的趋势下,集群呈现规模大、异构性强的特点;尤其是异构性比较明显的集群,节点的类型多种多样,节点的性能更是参差不齐,低性能节点的不恰当使用会严重制约Docker镜像分发的效率.为此,在本文中,我们基于实际应用场景的真实特性,提出了一种异构云环境下高效的Docker镜像分发方案,并实现成Docker镜像分发系统,记为RainbowD.RainbowD采用了一种新型的高并发机制处理镜像分发的任务;首先,发挥集群中所有节点的作用来减轻仓库节点的负载;其次,使用一种性能优先级排序的方法动态的选择性能高的节点,在有限的资源下提升镜像分发效率;最后,由于对节点进行优先级排序,集群中每个节点不需要再向其余所有节点发送消息,并首次使用SSH协议分发镜像,解决了P2P网络中存在的消息延迟、消息重复和冗余等问题,减少了带宽资源的浪费.特别地,由于BitTorrent协议的安全性的弊端备受学者诟病,考虑到云环境下集群的安全性问题,利用SSH协议加密特性和安全性验证,提升了云环境的安全性.在本文中,我们设置了两个独立实验,结果如下:首先,RainbowD和其他镜像分发方案对比验证其镜像分发效率,在实验的异构云环境下,RainbowD与使用Docket的方案相比,镜像分发效率显著提升,分发效率提高了30%以上,分发�In recent years,the rapid development of Internet technology has led to an exponential increase in the number of Internet users.As a result,the number of services required by users is increasing,and the types of services are increasing.Not only that,with the virtualization technology developing,Docker is favored by developers for its lightweight nature and good portability.Docker is playing an increasingly important role in virtualization technology.In order to improve the efficiency of service deployment,more and more people use Docker to deploy services in cloud environments.When deploying services in a cluster,you need to distribute images to nodes.Therefore,how to efficiently distribute Docker images has become a hot topic of research.However,under the development trend of the contemporary Internet,clusters are characterized by large scale and strong heterogeneity.Especially for clusters with relatively obvious heterogeneity,there are many types of nodes,and the performance of the nodes is uneven.So,the inappropriate use of low-performance nodes will seriously restrict the efficiency of Docker image distribution.To this end,in this paper,we propose an efficient Docker image distribution scheme in a heterogeneous cloud environment based on the real characteristics of actual application scenarios,and implement it as a Docker image distribution system,which is called as RainbowD.RainbowD adopts a new high-concurrency mechanism to handle the task of image distribution.First of all,it uses all the nodes in the cluster to release the load of the registry.Secondly,a performance-prioritized method is used to dynamically select nodes with high performance for improving the efficiency of Docker image distribution with limited resources in a heterogeneous cloud environment.Finally,because the nodes are prioritized,each node in the cluster does not need to send messages to all the other nodes,and RainbowD use the SSH protocol to distribute the images for the first time,which solves the disadvantages of message delay,mess

关 键 词：Docker技术 异构云环境 镜像分发 高并发 SSH协议 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]