节点文献

虚拟化计算系统性能评测管理工具的研究

Research on Virtual Machine Evaluation Management Facility for Virtualization Computing System

【作者】 陈青茶

【导师】 袁平鹏;

【作者基本信息】 华中科技大学 , 计算机系统结构, 2008, 硕士

【摘要】 随着计算系统的资源规模不断扩展、处理能力快速增强、资源种类日益丰富、应用需求灵活多样,寻求新型的计算机理论与模式已成为未来信息技术领域面临的重大挑战。计算系统虚拟化作为一种新型计算模式推动科技发展和社会进步,已成为国际竞争的焦点和制高点。然而,随着虚拟化技术大量的受到关注和虚拟化系统产品的不断推出,对于虚拟化系统性能的评测也成为急需解决的问题之一。传统计算机性能评测系统比如LMbench,Vbench, SPEC等,虽然可以应用与单个虚拟机的性能评测中。但是,在虚拟机系统之中,一个物理机上往往可以存在多个虚拟机。采用传统测试方法就会存在两个方面的问题:其一,多虚拟机测试时,虚拟机之间可能会产生相互干扰;其二,由于传统基准程序的测试过程比较繁琐,在对多机进行测试时,这个繁琐的过程将更加严重。虚拟机系统性能评测管理系统是一个半自动测试管理工具,它可以有效地解决这两个问题。该系统采用C/S模式,监控器端存储基准程序信息和测试信息,守护程序运行有虚拟机后台,管理基准程序的运行。为了支撑强大的功能,在系统中设计了序列化的数据包管理机制、基于自定义事件的核心处理机制、嵌入式数据库引擎和基准程序运行控制器等主要核心技术部件。辅助其他模块,最终实现了多虚拟机集中管理,基准程序自动部署、测试过程远程管理、多机同时测试、测试结果自动回收和分析、性能辅助分析的功能以及支持外部基准程序、支持Xen,VMware以及VirtualPC等几种主流虚拟机和支持Linux和Windows操作系统。为多虚拟机系统提供了十分便利的半自动测试手段,也满足了多虚拟机测试时的排除干扰的需求。通过在单机多虚拟机的真实环境中使用本系统进行lmbench的测试,测试的结果表明,本系统完全达到了上述功能要求,简化了繁琐的测试过程和数据分析过程。此外,在性能开销方面本系统也做到了达到了0.5%以下,系统延时间也限制在0.5s以内,0.5%的系统开销和0.5s的延迟,对于本系统来说,几乎可以忽略。

【Abstract】 With the continuous expanding of computing system resources and the demand for flexible applications, the demand for a new computing mode has been a major challenge in the field of information technology. Computing system virtualization, which is a new computing mode promoting science and technology development and social progress, has become the focus of international competition. However, with more and more company and institutes participated in and continuous new virtualization products were released, virtual computing system performance evaluation has become one of the issues that need to be resolved. Traditional computing system performance evaluation system such as lmbench, dbench, SPEC and etc. can be applied to evaluate single virtual machine. Usually, there will be multiple virtual machines in a virtualized system. Therefore, traditional evaluating method will give rise to two new problems: First, while multiple virtual machine testing, virtual machines may interfere with each other; second, multiple virtual machine evaluating can be rather boring for traditional benchmark method for a single machine is very complicated.Virtual computing system performance evaluation and supervision system(VEMF) is a semi-automatic benchmark supervision tool, which can resolve the two problems efficiently. VEMF which adopts C/S mode include a server, some clients and a supervising interface. The server stores benchmark information and tests information while the client monitors the local running benchmarks and communicates with the server. To support advanced functions, VEMF introduces some core component such as sequentialized data-packet management mechanism, user-defined event based core handling mechanism, embedded database engine, benchmark controller. With these core modules and other assistant modules, VEMF provides with these functions: benchmark auto-deploy, remote startup, test results auto-acquisition, assistant analysis and etc. It makes single-physical-multi-VM computing system performance evaluation much more convenient.Test results reveal that VEMF has a favorable performance and can simplify testing process data analysis process greatly. In addition, the additional cost and latency produced by VEMF are so minimum that can be ignored.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络