软件定义数据中心网络的虚拟机迁移方案研究

发布时间：2018-03-18 07:08

本文选题：数据中心　切入点：软件定义网络　出处：《山东大学》2017年硕士论文　论文类型：学位论文

【摘要】：现代数据中心利用虚拟化技术使得多个相互独立的虚拟机同时运行在一台物理服务器上,实现高效的资源管理,以降低成本和能源预算。在数据中心网络中,为了实现网络的优化管理目标(如节能、负载均衡等),一些虚拟机需要从虚拟机的初始位置迁移到目标位置。动态虚拟机迁移是一种可以实现数据中心优化管理目标的有效方法。当多个虚拟机同时需要迁移时,虚拟机的迁移顺序会极大的影响虚拟机的迁移性能;由于网络的带宽是有限的,为不同的虚拟机分配不同的迁移带宽也会对虚拟机的迁移性能带来较大的影响。因此需要制定一个有效的迁移方案来决定虚拟机的迁移顺序和迁移带宽的分配以提高虚拟机迁移性能。数据中心的规模正在不断的扩大。然而传统的网络技术使得虚拟化的程度严重不足。软件定义网络(Software Defined Networking,SDN)的出现可以有效的解决传统网络带来的虚拟化程度不足的问题,为虚拟机迁移问题提供了更加有效的解决方案。SDN的中心控制可使得网络实现多路径转发等多方面的要求。利用SDN的优势可以提高网络的传输能力以减少虚拟机总的迁移时间(Total Migration Time,TMT)和停机时间(Total Downtime,TD)。因此,在软件定义的数据中心网络中,为了制定有效的虚拟机迁移方案,本文提出了一个个体虚拟机迁移性保证的虚拟机迁移(Individual Performance Guaranteed VM Migration,IPG-VMM)算法和基于粒子群的虚拟机迁移(PSO-based VM Migration,PSO-VMM)算法。(1)IPG-VMM算法通过虚拟机迁移的预处理以及最小化不同虚拟机迁移方案之间的相关度来最大化总的迁移带宽,同时通过保证个体VM的最小迁移带宽来保证其迁移性能。基于IPG-VMM算法的虚拟机迁移方案能够解决逐步(step-by-step)迁移的"空闲带宽"问题,可以充分利用网络的带宽,提高网络带宽资源的利用率。(2)为了得到更好的虚拟机迁移方案,本文还使用了群智能算法。使用群智能算法可以通过牺牲一定的问题求解时间,来得到更好的解。虚拟机迁移问题是一个组合优化问题,粒子群算法对于解决组合优化问题是一个较为有效的方法,算法的优化性能也得到了证明。因此,本文通过合并粒子来改善算法中粒子的更新,并提出一个基于粒子群算法的虚拟机迁移方案,将对应的算法称为基于粒子群的虚拟机迁移(PSO-basedVM Migration,PSO-VMM)算法。(3)实验表明,相较于FPTA算法,IPG-VMM算法最多可以节省25%的TMT和50%的TD;而相较于FPTA算法,IPG-VMM算法最多可以节省65%的TMT和30%的TD。通过牺牲一定的问题求解时间,PSO-VMM算法制定的迁移方案在TMT和TD方面都优于IPG-VMM算法。
[Abstract]:Modern data centers use virtualization technology to enable multiple independent virtual machines to run on one physical server at the same time to achieve efficient resource management to reduce costs and energy budgets. In order to achieve the optimal management objectives of the network, such as energy saving, Load balancing, some virtual machines need to migrate from the initial location of the virtual machine to the target location. Dynamic virtual machine migration is an effective method to achieve the data center optimization management objectives. The migration order of the virtual machine greatly affects the migration performance of the virtual machine; because the bandwidth of the network is limited, Allocation of different migration bandwidth for different virtual machines will also have a great impact on the migration performance of virtual machines. Therefore, an effective migration scheme is needed to determine the migration order of virtual machines and the allocation of migration bandwidth. Improving the performance of virtual machine migration. The scale of data center is expanding. However, the traditional network technology makes the degree of virtualization seriously insufficient. The emergence of software definition network Defined networking SDN can effectively solve the problem of traditional network. The resulting problem of inadequate virtualization, It provides a more effective solution for virtual machine migration. The central control of SDNs enables the network to realize multipath forwarding and other requirements. Using the advantages of SDN can improve the transmission capacity of the network and reduce the total number of virtual machines. Total Migration time (TMTT) and downtime total download time (TDT). Therefore, In a software-defined data center network, in order to develop an effective migration scheme for virtual machines, In this paper, we propose an individual Performance Guaranteed migration algorithm with guaranteed mobility and a particle swarm based migration algorithm called PSO-based VM migration / PSO-VMMM. We propose an IPG-VMM algorithm based on the preprocessing of virtual machine migration and minimization of different virtual machines. The correlation between the proposed migration schemes to maximize the total migration bandwidth, At the same time, the migration performance is guaranteed by ensuring the minimum migration bandwidth of individual VM. The virtual machine migration scheme based on IPG-VMM algorithm can solve the "idle bandwidth" problem of stepwise step by step migration, and can make full use of the bandwidth of the network. In order to obtain a better migration scheme of virtual machine, this paper also uses swarm intelligence algorithm, which can solve the problem at the expense of certain time. Virtual machine migration problem is a combinatorial optimization problem, particle swarm optimization algorithm is a more effective method for solving combinatorial optimization problem, and the optimization performance of the algorithm is also proved. In this paper, particle updating in the algorithm is improved by merging particles, and a virtual machine migration scheme based on particle swarm optimization is proposed. The corresponding algorithm is called PSO-based VM migration (PSO-VMMM) algorithm. Compared with the FPTA algorithm, the IPG-VMM algorithm can save up to 25% TMT and 50% TDs, while the FPTA algorithm can save 65% TMT and 30% TD.The migration scheme developed by the PSO-VMM algorithm at the expense of some problem solving time in TMT. Both TD and TD are superior to IPG-VMM algorithm.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP302

【相似文献】