【作者】 张浩；

【导师】 李善平；

【作者基本信息】 浙江大学 ， 计算机技术， 2018， 硕士

【摘要】 随着多年来虚拟化技术不断发展,云计算使用虚拟化技术来实现大规模计算资源的池化管理和弹性伸缩。在过去十多年,Hypervisor技术一直在数据中心基础设施中扮演着骨干的角色,用于提供计算资源的隔离和管理。现代IT企业已经不再满足于仅通过IaaS获取具有计算能力的虚拟机节点,而PaaS又给应用的开发上引入了一些限制。基于容器即服务的云平台Kubernetes的出现为实现新型数据中心操作系统提供的了方便。Kubernetes默认使用的是基于操作系统级虚拟化技术的Docker容器引擎,容器与传统的虚拟机相比,额外资源开销更小,但在隔离性方面还有诸多问题。然而现代数据中心内往往承载着差异化的数据业务和复杂的应用程序,一个稳定、可靠、安全的容器引擎显得尤为重要。本文重点分析了目前主流容器技术的特性和应用服务在云平台的运行需求,在深入研究Kubernetes系统架构后,提出在Kubernetes中引入基于传统虚拟化技术的HyperContainer容器引擎和Unikernel技术用于承载容器化应用,以达到满足虚拟机级别甚至更高的隔离性要求和更多的应用场景覆盖。并将Docker容器引擎、HyperContainer容器引擎和Unikernel放在一起协同工作,在Kubernetes中设计一个全新的容器运行时中间层用于管理三种容器方案,将不同的工作负载发送到最适合的容器引擎中。最后为这套混合异构容器运行时分别设计了网络模块,能够支持CNI网络模型并可以共同使用一个网络配置。最后的实验数据表明本文设计的混合异构容器运行时能够承担Kubernetes官方要求的所有功能特性和工作负载。更多还原

【Abstract】 With the continuous development of virtualization technology over the years,cloud computing uses virtualization technology to achieve pooled management and flexible scalability of large-scale computing resources.For more than a decade,Hypervisor technology has played a central role in data center infrastructure to provide the isolation and management of computing resources.Modern IT companies are no longer content with acquiring compute-capable virtual machine nodes through IaaS only,and PaaS introduces some limitations to application development.The emergence of Kubemetes,a container-as-a-service platform,facilitates the implementation of new data center operating systems.By default,Kubernetes uses the Docker Container Engine,based on operating system-level virtualization technology,with extra overhead for containers compared to traditional virtual machines,but there are still many issues with isolation.However,modern data centers often carry a differentiated data services and complex applications,a stable,reliable and secure container engine is particularly important.This article focuses on the analysis of the current mainstream container technology features and application services running on the cloud platform needs,after in-depth study of Kubernetes system architecture,proposed in Kubernetes HyperContainer container technology based on traditional virtualization technology and Unikernel technology for container containerization Application,in order to meet the virtual machine level or even higher isolation requirements and more scenarios covered.Working with Docker Container Engine,HyperContainer Container Engine,and Unikernel,a new container runtime middleware in Kubernetes is designed to manage three container scenarios,sending different workloads to the most appropriate container engine.Finally,we designed the network modules for the hybrid heterogeneous container runtime respectively,which can support the CNI network model and can use a network configuration together.The final experimental data indicates that the hybrid heterogeneous container designed in this paper is capable of assuming all the functionalities and workloads officially required by Kubernetes.更多还原