虚拟化环境下的多机内存优化技术研究与实现

发布时间：2018-01-05 18:32

本文关键词：虚拟化环境下的多机内存优化技术研究与实现　出处：《首都师范大学》2013年硕士论文　论文类型：学位论文

【摘要】：随着计算机以及网络技术的发展,系统虚拟化技术越来越广泛的被数据中心所使用,通过在数据中心集群服务器上建立虚拟机集群,从而最大化利用现有的服务器硬件资源,同时降低数据中心的运营成本。部署虚拟机的关键在于如何在不影响系统性能的情况下在服务器上部署更多的虚拟机。然而限制服务器上部署更多虚拟机的主要瓶颈在于内存,因此如何让集群中虚拟机在不影响性能的情况下做到所需内存最小变得越来越重要。而现今条件下,系统虚拟化的研究重点大都集中宿主机为单机情况下,而对于单机状态下由于自身资源有限,所以优化的空间也非常有限,然而服务器集群内资源优化的潜力没有发掘出来。基于上述原因,本文提出了虚拟环境下一种多机状态下内存优化的解决方案。将集群中所有的物理内存通过逻辑扩展为一个大的地址空间,然后在该地址空间上建立动态内存池,通过修改虚拟机监控器(Virtual Machine Monitor,VMM)向上提供一个统一的内存管理接口。所有的虚拟机运行在这个内存空间上,这样就可以在整个集群上而非单机上对内存进行优化。通过构建地址空间映射组件,将动态内存池中的内存映射到集群中各服务器的物理地址空间上,并且对内存故障进行相应的处理。通过修改基于内存页面内容的内存共享方法,在全局地址空间内查找相同的内存页面,通过共享这些内存页面,达到节省更多内存的目的。同时,为了配合上述需求,对内存的访问加以优化,通过预取由多马尔科夫链模型预测的要访问的内存页面,加快访问速度,使用主动内存技术从而充分利用各服务器的计算资源。本文在KVM的原型系统中实现上述功能,并且针对上述目的做了相应的测试。实验结果表明,本文完成的相关工作可以有效的节省更多的内存,从而使得集群化虚拟机系统更高效。
[Abstract]:With the development of computer and network technology, system virtualization technology is more and more widely used by the data center, by building virtual machine cluster on the data center cluster server. To maximize the use of existing server hardware resources. The key to deploying virtual machines is how to deploy more virtual machines on the server without affecting system performance. However, the main constraints to deploying more virtual machines on the server are. The bottleneck is memory. Therefore, it is more and more important to make the virtual machine in the cluster to minimize the required memory without affecting the performance. However, under the present conditions, the research focus of system virtualization is mostly concentrated on the single host. However, the potential of resource optimization in the server cluster is not explored because of the limited resources in the single machine state, so the space for optimization is also very limited. For the above reasons, this paper proposes a solution of memory optimization in virtual environment, which expands all the physical memory in the cluster into a large address space through logic. The dynamic memory pool is then created on the address space, and virtual Machine Monitor is modified by modifying the virtual machine monitor. VMMs provide a unified memory management interface. All virtual machines run on this memory space. By building address space mapping components, memory in the dynamic memory pool can be mapped to the physical address space of the servers in the cluster. By modifying the memory sharing method based on the contents of memory pages, we can find the same memory pages in the global address space and share these memory pages. To achieve the purpose of saving more memory. At the same time, in order to meet the above requirements, the memory access is optimized, and the access speed is accelerated by prefetching the memory pages to be accessed predicted by the multi-Markov chain model. Use active memory technology to make full use of the computing resources of each server. This paper implements the above functions in the prototype system of KVM, and makes a corresponding test for the above purpose. The experimental results show that the related work can save more memory effectively. Therefore, the cluster virtual machine system is more efficient.
【学位授予单位】：首都师范大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333

【参考文献】