基于虚拟计算环境的内存资源弹性分配研究

发布时间：2018-01-03 06:38

  本文关键词：基于虚拟计算环境的内存资源弹性分配研究　出处：《国防科学技术大学》2013年硕士论文　论文类型：学位论文

  更多相关文章： 虚拟计算环境 虚拟化技术 内存资源 内存空洞 换页优化

【摘要】：社会生活的各个领域对互联网提出的新的迫切需求和大量新型网络应用的出现,给网络中的海量资源的有效管理带来了新的挑战。特别是互联网中的资源具有成长性、自治性和多样性特点,使得它们在协同共享和综合利用的过程中出现了诸多新的问题。其中,广域分布和异构的内存资源共享和分配这一实际问题,成为其中重要的子问题。作为三大计算资源之一,内存资源相对于CPU和I/O资源更难被共享和复用,因此对内存资源弹性分配和共享的研究工作具有重要的学术意义和商业价值,尤其是在新的互联网应用下,内存使用场景将变得更加不可预知。围绕这一问题,本文主要对以下四项内容进行了深入的研究:一、深入分析虚拟化平台和XEN体系结构,以及半虚拟化驱动,内存虚拟化模型等技术,它们是实现虚拟化环境中内存整合的关键支撑技术;接着,本文又重点研究了内存气球方法、内存热插拔技术、页共享技术和在线迁移技术等,在分析了它们的优点的同时,也指出了各自的不足之处;二、针对使用气球方法的虚拟机在内存变化幅度较大时可能会出现性能退化的情形,我们提出了内存空洞方法,以达到解决该问题的目的。内存空洞在虚拟机监视器中分配不具有物理内存的“空洞区域”,当虚拟机对该内存进行访存操作时,才分配真实的物理内存。实验证明内存空洞方法确实能够在内存激增的情况下,合理分配内存,保证虚拟机服务质量;三、对客户虚拟机换页问题的深入探索后,我们提出了基于宿主机的换页缓存池方法,来对客户机的换页操作进行优化。该方法利用了宿主机中的部分内存来合理缓存由客户机中换出的部分内存页面,并依照每台虚拟机内存资源使用情况合理地分配缓存池大小。缓存池方法利用宿主机中的资源为客户机提供换页服务,既平衡了宿主机和客户机内存使用量,又提高了整体的内存利用效率;四、进一步地,我们提出了基于网络内存的客户机换页方法,以克服单个物理节点内存使用的上限。该方法利用其他物理节点上的内存资源,有效组织成为一个换页硬盘,用于存储客户机换出的内存页面。在使用了先进的网络设备(如Infiniband或者10Gb以太网)后,经过测试,对于传统换页方法,基于网络内存方法在换页性能上有一定的优势。本文是对虚拟计算环境中虚拟化平台内存资源弹性分配的一次有益探索,研究成果对于搭建灵活易扩展并拥有海量数据处理能力的云计算平台具有良好的理论价值和实践意义。本文所做的工作已在承研的国家重点基础研究发展计划(973计划)、国家高技术研究发展计划(863计划)和自然科学基金中得到了应用。
[Abstract]:Various fields of social life put forward a new urgent need for the Internet and the emergence of a large number of new network applications. It brings new challenges to the effective management of the massive resources in the network. Especially, the resources in the Internet have the characteristics of growth, autonomy and diversity. In the process of collaborative sharing and comprehensive utilization, there are many new problems, including wide area distribution and heterogeneous memory resource sharing and allocation. As one of the three computing resources, memory resources are more difficult to share and reuse than CPU and I / O resources. Therefore, the research on the flexible allocation and sharing of memory resources has important academic significance and commercial value, especially in the new Internet applications, memory usage scenarios will become more unpredictable. This paper mainly studies the following four contents: first, in-depth analysis of virtualization platform and XEN architecture, as well as paravirtualization driver, memory virtualization model and other technologies. They are the key supporting technologies to realize memory integration in virtualized environment. Then, this paper focuses on the memory balloon method, memory hot-plug technology, page sharing technology and online migration technology, while analyzing their advantages, but also pointed out their shortcomings; Secondly, we propose a memory void method to solve the problem that the performance of virtual machine with balloon method may deteriorate when the memory change is large. To solve this problem, memory holes allocate "empty areas" that do not have physical memory in the virtual machine monitor, when the virtual machine accesses the memory. The experiment proves that the memory empty method can allocate memory reasonably and guarantee the quality of service of virtual machine in the case of large increase of memory. Thirdly, after deeply exploring the problem of customer virtual machine page changing, we propose a new method of page changing cache pool based on host computer. This method utilizes part of memory in the host to reasonably cache some memory pages swapped out of the client. According to the memory resource usage of each virtual machine, the buffer pool can reasonably allocate the buffer pool size. The cache pool method uses the resources in the host computer to provide page changing service for the client, which balances the memory usage between the host computer and the client. It also improves the efficiency of memory utilization. Fourth, we propose a network memory-based client paging method to overcome the upper limit of memory usage of a single physical node, which utilizes memory resources on other physical nodes. Effectively organized as a paging disk for storing memory pages swapped out by the client. Tested after using advanced network devices such as Infiniband or 10GB Ethernet. For the traditional paging method, the network-based memory method has some advantages in the performance of page wrapping. This paper is a useful exploration for the flexible allocation of memory resources in virtual computing environment. The research results have good theoretical value and practical significance for building a cloud computing platform that is flexible and scalable and has huge data processing capability. The work done in this paper has been studied in the national key basic research and development plan (. 973). The National High-tech Research and Development Program (NRDP) and the Natural Science Foundation have been applied.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333.1
，

本文编号：1372830