虚拟化云平台下内存资源按需分配与协同调度方法的研究

发布时间：2018-01-22 04:56

本文关键词： 云计算虚拟化内存资源按需分配透明感知协同调度　出处：《国防科学技术大学》2013年硕士论文　论文类型：学位论文

【摘要】：近年来,社交网络、电子商务、在线视频、大容量网盘等新一代互联网应用发展迅猛,这些应用具有数据规模大、业务增长快等特点,需要越来越多的高性能服务器支持,软硬件维护成本高昂。云计算正是在这样的背景下诞生的,云计算整合以数据中心的形式出现的大规模计算、存储和网络资源,以低廉的成本为互联网用户提供多种类型的服务。用户可以根据业务负载快速申请和释放资源,并以按需支付的方式对所用的资源付费,在保证服务质量的同时大大降低成本。虚拟化技术将计算、存储、网络等资源抽象化并统一表示,是构将云平台的技术基础。相比传统技术,虚拟化云平台的优势在于资源的弹性配置,可是由于“语义鸿沟”的存在,内存资源的弹性配置一直难以实现。为解决内存弹性配置这个难题,本文围绕着内存资源的按需分配和协同调度进行了研究,主要研究内容包括:一、为实现对虚拟机内部内存使用情况实时而透明的感知,本文在深入分析了内存虚拟化技术的原理和特点的基础上,在不需修改虚拟机操作系统的前提下,提出并实现了x86模式下基于进程和页表可见性的内存资源使用状况的感知方法,相比原有其他方法具有开销低、准确度高、实时性好的特点。二、在深入分析了内存与存储之间的协同作用原理的基础上,本文提出了内存与存储协同的资源分配方法,在资源分配的过程中,充分考虑了缓存的需求和换页分区的使用。通过监控虚拟机的磁盘访问序列,发现数据重用并计算重用距离,根据重用距离来判断缓存的紧张程度,并计算合理的缓存分配量。本方法大大提高了I/O密集型应用的性能。三、在虚拟机内存使用情况透明感知和内存与存储之间的内存资源分配方法的基础之上,本文进一步提出了兼顾公平与优先级的多虚拟机间的内存资源调度策略,该策略能在公平地保证虚拟机运行性能的前提下,兼顾对虚拟机优先级的差异化配置,提高内存资源的利用率和效能。本文是对虚拟化云平台上内存资源按需分配和协同调度的一次有益探索,研究成果对于搭建灵活易扩展、高效能的云计算平台具有良好的理论和实践价值。本文所做的工作已在承研的与云计算和虚拟计算环境相关的973项目和863项目中得到了应用。
[Abstract]:In recent years, the new generation of Internet applications, such as social network, electronic commerce, online video, mass network disk and so on, have developed rapidly. These applications have the characteristics of large scale of data and fast growth of business. More and more high-performance servers are needed, and the maintenance cost of hardware and software is high. Cloud computing is born in this context, cloud computing integration in the form of data center of large-scale computing. Storage and network resources provide Internet users with various types of services at low cost. Users can quickly apply for and release resources according to the business load and pay for the resources on demand. Virtualization technology abstracts computing, storage, network and other resources into a unified representation, which is the technical basis of constructing cloud platform. Compared with traditional technology, virtualization technology can greatly reduce the cost. The advantage of virtualization cloud platform lies in the flexible allocation of resources, but due to the "semantic gap", the flexible allocation of memory resources has been difficult to achieve. This paper focuses on the allocation of memory resources on demand and collaborative scheduling. The main research contents are as follows: first, to realize real-time and transparent perception of memory usage in virtual machine. Based on the deep analysis of the principle and characteristics of memory virtualization technology, this paper does not need to modify the virtual machine operating system. This paper proposes and implements the memory resource usage perception method based on process and page table visibility in x86 mode. Compared with other methods, it has the characteristics of low overhead, high accuracy and good real-time performance. On the basis of deeply analyzing the principle of cooperation between memory and storage, this paper proposes a resource allocation method of memory and storage coordination, which is used in the process of resource allocation. By monitoring the disk access sequence of virtual machine, we can find the data reuse and calculate the reuse distance, and judge the shortage of cache according to the reuse distance. And calculate the reasonable cache allocation. This method greatly improves the performance of I / O intensive applications. 3. Based on the transparent perception of virtual machine memory usage and the memory resource allocation method between memory and storage, this paper proposes a memory resource scheduling strategy which considers fairness and priority. The strategy can guarantee the performance of the virtual machine fairly and give consideration to the priority of the virtual machine. Improve the utilization and efficiency of memory resources. This paper is a useful exploration of memory resource allocation and collaborative scheduling on virtualized cloud platform. The research results are flexible and easy to expand. High performance cloud computing platform has good theoretical and practical value. The work done in this paper has been applied in 973 projects and 863 projects related to cloud computing and virtual computing environment.
【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

【相似文献】