云计算虚拟化平台的内存资源全局优化研究

发布时间：2018-03-24 11:15

本文选题：虚拟化　切入点：内存资源　出处：《东北大学》2013年硕士论文

【摘要】：随着互联网的飞速发展,其信息服务的快速增长以及用户对信息服务的逐渐依赖的趋势已经势不可挡了。而互联网服务提供商一直以来,面临着以更低成本提供更好服务的挑战。因此拥有按使用付费和高性价比的云计算模式受到了越来越多企业以及用户的青睐。而借助虚拟化技术,云计算能将大规模计算资源统一管理,提高了资源利用效率,简化了管理和维护成本,并为用户提供易获取、易扩展的按需服务。然而虚拟化技术和云计算平台的结合带来了全新的使用模式和资源整合,基于虚拟化技术的资源按需分配与调度可以提高云平台资源的利用率,并且提升云服务的服务质量,甚至它降低了云用户的总体拥有成本。但是,物理服务器的资源边界限制了资源的全局优化能力。而现有的研究是在云平台对各个虚拟机之间的内存资源进行调度算法的优化,从而提高整体的内存资源利用率。或者是进行架构方面的设计,从地址映射方面,页面交换机制,空闲页面回收方面来进行改进,从而达到资源利用的目的。这些方法虽然都可以很好地提高内存资源利用率,但大部分的虚拟机的内存资源无论是空闲还是紧张,都需要重新进行一次内存的分配、调度,或者是映射。这在无形中就会增加虚拟机的负担,从而会相对地降低内存资源利用率。本文在其已有的全局优化框架上进行相应的改进,增添了当虚拟机内部资源空闲时的最小内存边界值,当虚拟机内存资源利用率低的时候,需要比较最小内存边界值,再将多余的内存资源映射到全局的空闲内存池中。然后基于上述框架,将虚拟机的内存资源再细分为利用率低和利用率高的情况进行研究,并分别给出了两种调节算法,以及这两种算法之间的相互关系。当内存利用率低的时候,虚拟机将先比较判断最小边界值,将部分空闲内存放入的全局的空闲内存池中。而当虚拟机内存资源利用率高时,将通过全局空闲内存池进行全局调节。结果是既降低了每次与全局空闲内存池交换的次数,又降低了虚拟机之间的内存交换次数,平均内存资源利用效率将大大提高。最后本文对研究的内容进行了实验分析。实验结果表明,该框架以及算法能够很好地优化云平台中内存资源配置,提升整个平台的资源利用率,使关键任务的执行有显著的加速。测试结果更加展示了本文方法的性能优势以及良好的可用性。
[Abstract]:With the rapid development of the Internet, the rapid growth of its information services and the gradual dependence of users on information services have become unstoppable. Faced with the challenge of providing better services at lower cost, cloud computing with a pay-per-use and cost-effective model is increasingly popular among businesses and users. Cloud computing can unify the management of large-scale computing resources, improve resource utilization efficiency, simplify management and maintenance costs, and provide easy access to users. However, the combination of virtualization technology and cloud computing platform brings new usage patterns and resource integration. Resource allocation and scheduling based on virtualization technology can improve the utilization of cloud platform resources. And to improve the quality of service of cloud services, even reducing the overall cost of ownership of cloud users. However, The resource boundary of the physical server limits the global optimization ability of the resource, and the existing research is to optimize the memory resources among the virtual machines in the cloud platform. In order to improve the overall utilization of memory resources. Or the design of architecture, from address mapping, page exchange mechanism, free page recovery to improve, In order to achieve the purpose of resource utilization, although these methods can improve the utilization of memory resources, most of the memory resources of virtual machine, whether idle or tight, need to be allocated and scheduled again. Or mapping. This will increase the burden of virtual machines, which will reduce the utilization of memory resources. In this paper, the existing global optimization framework is improved. Adds the minimum memory boundary value when the virtual machine internal resource is idle, and the minimum memory boundary value needs to be compared when the virtual machine memory resource utilization is low, Then the redundant memory resources are mapped to the global free memory pool. Then, based on the above framework, the memory resources of the virtual machine are subdivided into low utilization and high utilization, and two adjustment algorithms are given respectively. When the memory utilization is low, the virtual machine will compare and judge the minimum boundary value, and put part of the free memory into the global free memory pool. When the memory utilization of the virtual machine is high, the virtual machine will put part of the free memory into the global free memory pool. The global adjustment will be made through the global free memory pool. The result is that the number of exchanges with the global free memory pool is reduced each time, and the number of memory exchanges between virtual machines is also reduced. The average efficiency of memory resource utilization will be greatly improved. Finally, the experimental results show that the proposed framework and algorithm can optimize the allocation of memory resources in cloud platform. Improving the resource utilization of the whole platform can accelerate the execution of critical tasks significantly. The test results show the performance advantages and good usability of this method.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP393.09

【参考文献】