基于SLA的私有云资源管理框架的研究与实现

发布时间：2018-07-01 21:05

本文选题：私有云 + SLA　；参考：《上海交通大学》2013年硕士论文

【摘要】：云计算环境下的资源管理主要包括计算资源、存储资源和网络资源等的管理。私有云是云计算的一种部署模式，与公有云相比，私有云的计算资源相对有限，其计算资源，特别是虚拟机资源的管理问题尤为突出。本文工作是以某省交通物流云计算平台建设的实际项目为背景，根据该项目的实际需求，其私有云计算环境的虚拟机资源管理存在以下问题：1）虚拟机资源的有限性与应用规模的可扩展性存在矛盾的情况下，如何根据与用户签订的服务水平协议（Service LevelAgreement，SLA）既保证服务水平又合理分配虚拟机资源；2）为了充分利用资源，多个应用按照SLA分配的虚拟机资源位于同一台服务器上而且虚拟机资源的总和超过了服务器的物理资源上限时，如果某应用的负载增加导致抢占其他应用的资源，如何解决位于同一台服务器上的多个应用之间的性能隔离问题。针对上述问题，本文在对实际需求和相关技术深入分析的基础上，提出了基于请求准入控制的虚拟机资源分配方案以及基于预测和动态规划的虚拟机性能隔离方案。设计了一个基于SLA的虚拟机资源请求准入控制算法和一个基于预测和动态规划的虚拟机性能隔离算法。在此基础上，本文进一步阐述了一个基于SLA的私有云资源管理框架RMFPC (Resource Management Framework for Private Cloud)的设计与实现。测试及应用情况表明，该框架是可行及有效的。本文的主要研究内容包括： 1）针对私有云环境中基于SLA的虚拟机资源的合理分配问题，，提出了基于请求准入控制的虚拟机资源分配方案。该方案通过比较应用的SLA和主机空闲资源来决定是否分配虚拟机资源，其核心是基于SLA的虚拟机资源请求准入控制算法。该算法包括两部分，一是对满足条件的请求直接分配虚拟机资源，不满足条件的放入资源请求等待队列；二是响应应用退出事件，从等待队列中合理地选择请求并分配虚拟机资源。 2）针对同一台服务器上的多个应用之间的性能隔离问题，提出了基于预测和动态规划的虚拟机性能隔离方案，该方案的核心是基于预测和动态规划的虚拟机性能隔离算法。该算法通过曲线拟合方式预测虚拟机的资源抢占，从而决定是否迁移；然后通过动态规划算法确定迁移的目标主机。 3）在以上工作的基础上，设计并实现了一个基于SLA的私有云资源管理框架RMFPC。该框架除实现上述两个算法的核心模块外，还包括主机监视接口、虚拟机监视接口和虚拟机操作接口，这些接口的具体实现可适配目前主流的虚拟机管理软件商业产品。 4）对框架进行了测试，包括功能性测试和性能测试。虚拟机资源请求准入控制的功能性测试是通过四个模拟应用来完成；虚拟机性能隔离的功能性测试是通过增加一个应用的负载来进行；测试结果表明，框架实现了上述核心功能。本文模拟了一个虚拟机资源争用的场景，分别在有预测和虚拟机迁移、只有虚拟机迁移和没有预测也没有虚拟机迁移三种情况下，对框架的性能进行了测试，测试结果表明，同一个应用的运行时间第一种情况比第二种情况减少了约28%，比第三种情况减少了约50%。
[Abstract]:Resource management in the cloud computing environment mainly includes the management of computing resources, storage resources and network resources. The private cloud is a deployment model of cloud computing. Compared with the public cloud, the computing resources of private clouds are relatively limited. The computing resources, especially the virtual machine resources, are particularly prominent. This paper is based on the traffic in a province. The actual project of the cloud computing platform is the background. According to the actual requirements of the project, the virtual machine resource management of its private cloud computing environment has the following problems: 1) how to establish a service level protocol (Service LevelAgre) based on the limitation of the virtual machine resources and the scalability of the application scale Ement, SLA) ensures both the level of service and the rational allocation of virtual machine resources; 2) in order to make full use of resources, multiple applications of virtual machine resources allocated according to SLA are located on the same server and the sum of virtual machine resources exceeds the upper limit of the physical resource of the server, if the load increase of an application leads to the resource of preemption of other applications How to solve the problem of performance isolation between multiple applications on the same server.
In view of the above problems, this paper puts forward a virtual machine resource allocation scheme based on request admission control and a performance isolation scheme based on prediction and dynamic programming, and designs a SLA based virtual machine source request admission control algorithm and a prediction based on prediction. On the basis of this, the design and implementation of a private cloud resource management framework (Resource Management Framework for Private Cloud) based on SLA is further elaborated on this basis. The test and application show that the framework is feasible and effective.
The main contents of this paper are as follows:
1) aiming at the rational allocation problem of virtual machine resource based on SLA in private cloud environment, a virtual machine resource allocation scheme based on request admission control is proposed. This scheme determines whether the virtual machine resources are allocated by using the SLA and the idle resources of the host. The core is the SLA based virtual machine resource request admission control algorithm. The algorithm consists of two parts: one is to allocate the virtual machine resources directly to the request to satisfy the condition, and the resource request waiting queue is not satisfied. The two is to respond to the application exit event, select the request and allocate the virtual machine resources reasonably from the waiting queue.
2) aiming at the problem of performance isolation between multiple applications on the same server, a virtual machine performance isolation scheme based on prediction and dynamic programming is proposed. The core of the scheme is the performance isolation algorithm based on prediction and dynamic programming. The algorithm is determined by the curve fitting formula to predict the resource preemption of virtual machines. No migration is done; then the migrating target host is determined by dynamic programming algorithm.
3) on the basis of the above work, a SLA based private cloud resource management framework (RMFPC.) is designed and implemented. In addition to realizing the core modules of the above two algorithms, the framework also includes the host monitoring interface, the virtual machine monitoring interface and the virtual machine operation interface. The concrete implementation of these interfaces can match the current mainstream virtual machine management soft. A commercial product.
4) the framework is tested, including functional testing and performance testing. The functional testing of virtual machine resource request access control is completed through four simulation applications; the functional testing of the virtual machine performance isolation is carried out by adding an application load; test results show that the framework implements the above core functions. In this paper, a virtual machine resource contention scene is simulated, and the performance of the framework is tested with only three cases of prediction and virtual machine migration, only virtual machine migration and no prediction without virtual machine migration. The test results show that the first case of the same application is reduced by about 28% compared with the second cases. Third cases are reduced by about 50%.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP302

【相似文献】