云计算中多虚拟机交互的虚拟机部署位置选择策略研究

发布时间：2018-09-03 15:23

【摘要】：云计算以其低廉的硬件设备服务,高质量的服务体验,大大降低了企业购买大型服务器的成本和相应的维护费用。这也使得越来越多的企业将自身的多种业务迁移到云计算平台。企业通过云计算平台提供的基础设施即服务Iaa S,按需提出自己的资源请求。云中心根据企业的需求为企业分配足够的资源供其使用。Iaa S以虚拟机,即云服务器的形式,向用户提供资源。云中心控制器为这些云服务器选择合适的物理主机,作为放置云服务器的目标主机,这就是虚拟机部署位置选择问题。同时企业下辖业务之间并不是相互独立,业务之间是相互支撑协同完成任务,这些相交互云服务器构造了虚拟局域网。为这些云服务器选择目标主机位置,这就是本文研究的多虚拟机交互的虚拟机部署位置选择策略问题。选择最佳用于放置虚拟机的目标主机即为虚拟机部署位置选择策略的目标。该问题是一个多维装箱问题,是NP难问题,也是选择优化问题。本文从优化问题角度对问题进行分析求解。选择最小化电能消耗、最小化资源损耗作为其中两个目标,由于虚拟机之间组成了虚拟局域网,还要兼顾最小化网络通信开销,组成一个多目标优化问题。考虑到启发式群智能算法在解决多目标优化问题方面有着巨大的优势,本文将从群智能算法的角度给出研究问题的详细解过程。为实现快速高效的多虚拟机交互的虚拟机部署,同时折中实现对多个目标的优化,本文算法首先对虚拟机请求按照交互关系进行合理化分组,将资源请求重新组合成多组虚拟机集合请求,然后使用引入变异算子的标准粒子群优化算法,以适应度函数为度量值,搜寻最优放置方案。分组对虚拟机请求集合进行求解,这样避免了其他虚拟机请求对当前虚拟机集合产生的不利影响,能够更加合理的为虚拟机分配资源。变异算子避免了局部极值。适应度函数综合考虑了能耗、资源利用和虚拟机间交互产生的通信开销。最终提出本文研究的算法。为实现多虚拟机交互的虚拟机位置选择方案,采用以Cloud Sim云平台仿真平台为基础,Fat-Tree网络拓扑作为数据中心的底层架构,实现本文提出的启发式分组变异粒子群算法。该算法不仅能够避免局部极值的情况,还能避免不同虚拟机集合间在资源分配时产生的相互影响。同时使用SDN作为云平台的网络架构,测试不同的策略下虚拟机交互的时间延迟。为此,本文研究内容如下:首先,研究Cloud Sim模拟器,掌握平台的基本结构和功能,为编写模拟程序打下基础。其次,研究相关的群智能优化算法,特别是粒子群优化算法以及相应的改进算法。然后研究SDN的基本原理、控制模块以及Fat Tree拓扑结构。最后,在Cloud Sim平台实现粒子群算法,同时使用虚拟网络环境测试网络延迟。在上述工作的基础上,实现本文提出的启发式分组变异粒子群算法。根据云平台可用剩余资源,为多虚拟机分配目标主机,保证通信开销最小、资源利用率最大化,然后在虚拟网络环境下验证不同策略下的网络延迟。通过测试验证,启发式分组变异粒子群算法,保证了网络开销最小和全局最优解。
[Abstract]:Cloud computing greatly reduces the cost of purchasing large-scale servers and the corresponding maintenance costs because of its low-cost hardware services and high-quality service experience. The cloud center allocates sufficient resources for the enterprise to use according to its needs. Iaa S provides resources to users in the form of virtual machines, namely cloud servers. At the same time, the business under the jurisdiction of the enterprise is not mutually independent, and the business is mutually supporting and cooperating to complete the task. These interactive cloud servers construct a virtual local area network. Choosing the best target host for placement of the virtual machine is the goal of the VM deployment location selection strategy. This problem is a multi-dimensional packing problem, NP-hard problem, and also a selection optimization problem. In order to solve the multi-objective optimization problem, the heuristic swarm intelligence algorithm has great advantages in solving the multi-objective optimization problem. In this paper, the detailed solution process of the problem is given from the point of view of swarm intelligence algorithm. Rapid and efficient deployment of virtual machines with multiple virtual machine interactions and compromise optimization of multiple objectives are achieved. Firstly, the virtual machine requests are grouped rationally according to the interaction relationship, and the resource requests are re-grouped into multiple sets of virtual machine requests. Then, the standard particle swarm optimization algorithm with mutation operator is used to adapt to the requirements. The degree function is a measure and searches for the optimal placement scheme. The group solves the virtual machine request set, thus avoiding the adverse effects of other virtual machine requests on the current virtual machine set, and allocating resources for the virtual machine more reasonably. In order to implement the virtual machine location selection scheme for multi-virtual machine interaction, a heuristic packet mutation particle swarm optimization algorithm based on Cloud Sim cloud platform and Fat-Tree network topology as the underlying architecture of data center is proposed. This algorithm can not only avoid local extremum, but also avoid the interaction between different virtual machine sets in resource allocation. At the same time, SDN is used as the network architecture of cloud platform to test the time delay of virtual machine interaction under different strategies. The basic structure and function of the platform lay the foundation for writing simulation program. Secondly, the related swarm intelligence optimization algorithm, especially the particle swarm optimization algorithm and the corresponding improved algorithm are studied. Then the basic principle of SDN, the control module and the Fat Tree topology structure are studied. Finally, the particle swarm optimization algorithm is implemented on Cloud Sim platform, and the virtual algorithm is used. Based on the above work, a heuristic packet mutation particle swarm optimization algorithm is proposed. The target hosts are allocated to multiple virtual machines according to the remaining resources available in the cloud platform to ensure minimum communication overhead and maximize resource utilization. Then the network delay under different strategies is verified in the virtual network environment. Later, the heuristic packet mutation particle swarm optimization algorithm guarantees the minimum network overhead and the global optimal solution.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP302

【相似文献】