虚拟机动态资源分配及放置算法研究

发布时间：2018-01-07 02:20

本文关键词：虚拟机动态资源分配及放置算法研究　出处：《复旦大学》2012年硕士论文　论文类型：学位论文

【摘要】：“云计算”的概念自被提出之后,迅速成为科技领域最令人振奋的研究热点,受到工业界和学术界广泛关注。作为一项新兴的战略技术,虽然还没有一个关于它的统一定义,但学术界和工业界一般认为云计算是以虚拟化技术为基础、以按需付费为商业模式,具备弹性扩展、动态分配和资源共享等特点的新型网络化计算模式。其中基础设施即服务(Iaas)是云计算中最为基础及支撑性的服务,其存在的问题也是本论文集中关注的。IaaS通过虚拟化技术共享物理机资源池,用户通过向运营商租赁虚拟机来承载应用系统。随着运营商的业务发展,必将导致云IDC中的基础资源(如服务器,网络,存储等)的大量聚集,物理资源分配调度技术的优劣将直接影响到整体物理资源池的利用率,服务能力及SLA(Service-LevelAgreement,服务等级协议),因此成为IaaS云计算需要重点优化和突破的关键技术问题。对这些新问题的研究的缺乏将直接制约云计算资源的有效和集约的利用,降低云计算SLA的水平,造成云计算资源浪费及损失。传统的弹性云服务虽然在一定程度上能实现根据应用负载动态地对应用的资源进行增减,但通常是以虚拟机为单位的粗粒度的方法,造成资源利用率不足。另外,传统的虚拟机放置问题研究主要集中于资源利用的提高,而没有过多考虑多层应用之间的关联性。从而也会导致迁移一个过载的虚拟机之后给整个网络带来额外负载。本论文深入研究了现有的虚拟机资源分配模型及虚拟机放置算法,提出了基于负载预测的动态资源分配和应用感知的虚拟机放置算法。基于负载预测的动态资源分配算法通过对每台物理机中的虚拟机负载进行监测,并预估一段时间的资源使用情况,根据预测结果动态的在物理机层级为承载的虚拟机调整资源分配,因此能有效的、细粒度的使用物理机资源,减少全局迁移的发生。应用感知的虚拟机放置算法采用分组遗传算法,考虑不同虚拟机之间的耦合性,整合全局虚拟机的放置情况,减少物理机的使用,节约能耗,减少运营成本。
[Abstract]:Since the concept of "cloud computing" was put forward, it has rapidly become the most exciting research hotspot in the field of science and technology, and has been widely concerned by industry and academia. It is a new strategic technology. Although there is no uniform definition of it, academia and industry generally believe that cloud computing is based on virtualization technology, with a business model based on demand, and flexible expansion. A new network computing model with the characteristics of dynamic allocation and resource sharing. Infrastructure as a service (Iaas) is the most basic and supporting services in cloud computing, and its problems are also concerned in this collection. IaaS sharing physical computer resource pool through virtualization technology. Users rent virtual machines from operators to host application systems. With the development of operators' services, the basic resources (such as servers, networks, storage, etc.) in the cloud IDC will be aggregated. Physical resource allocation and scheduling technology will directly affect the overall physical resource pool utilization, service capacity and SLA(Service-LevelAgreement. Service level agreements (SLAs) have become the key technical issues for IaaS cloud computing to be optimized and broken through. The lack of research on these new problems will directly restrict the efficient and intensive utilization of cloud computing resources and reduce the level of cloud computing SLA. The traditional flexible cloud services can dynamically increase or decrease the application resources according to the application load to a certain extent. However, it is usually coarse-grained method based on virtual machine, which results in insufficient utilization of resources. In addition, the traditional research on virtual machine placement mainly focuses on the improvement of resource utilization. It also leads to the migration of an overloaded virtual machine and brings additional load to the entire network without too much consideration of the correlation between the multi-tier applications. In this paper, the existing virtual machine resource allocation model and virtual machine placement algorithm are deeply studied. A dynamic resource allocation algorithm based on load prediction and an application-aware virtual machine placement algorithm are proposed. The dynamic resource allocation algorithm based on load prediction monitors the load of virtual machine in each physical machine. And predict the use of resources for a period of time, according to the results of the prediction dynamically in the physical machine level for the load of the virtual machine to adjust the allocation of resources, so it can be effective, fine-grained use of physical computer resources. In order to reduce the occurrence of global migration, the perceptual virtual machine placement algorithm uses grouping genetic algorithm, considering the coupling between different virtual machines, integrating the placement of global virtual machines, and reducing the use of physical machines. Save energy consumption, reduce operating costs.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP302

【参考文献】