基于智能SDN的CDN用户请求分配优化机制研究

发布时间：2018-06-17 23:40

本文选题：软件定义网络 + 内容分发网络　；参考：《浙江大学》2017年硕士论文

【摘要】：近年来,信息技术迅猛发展,随着各种智能终端、应用的出现和宽带用户量的迅速增长,网络流量呈爆炸式增长,人们对于高质量的内容资源的需求也日益增加,给网络基础设施带来了巨大的压力。内容分发网络(Content Delivery Network,CDN)是一种提供可靠、有效的内容传输的技术。但是,传统的CDN无法获取网络全局信息,且依赖于DNS进行请求重定向,从而导致缺乏对路径和服务器选择的动态控制。软件定义网络(Software Defined Network,SDN)是一种新型的网络技术,它将控制面与数据面分离,从而提供灵活的集中式动态管理。将SDN技术应用于CDN网络中,可以实时监测网络状态信息,并及时作出决策,同时,可以在更细的时间细粒度上对用户请求进行重定向。论文首先提出了一种基于智能SDN的CDN网络架构。智能SDN通过引入智能中心,有效解决了对SDN控制器的功能需求不断增加、智能决策要求不断提升和多域SDN控制器间信息共享和交互等问题。论文提出了基于模型预测控制(Model Predictive Control,MPC)的CDN用户请求分配算法。主要根据用户QoE相关的两个核心参数:响应时间和用户带宽满足度,来对候选服务器和路径选择进行联合优化。仿真结果表明,该算法在响应时间和带宽满足度方面都有明显的性能提升。并发现可以通过调整权重参数来体现响应时间和带宽满足度两个参数的不同优先级,以满足不同网络应用的需要。为了进一步优化性能,论文又提出了一种将神经网络和MPC相结合的CDN用户请求分配算法。该算法首先基于神经网络对不同路径进行评分,并根据路径评分进行用户请求分配。仿真结果表明,由于神经网络的非线性映射能力和自学习能力,算法的性能得到了进一步的提升。
[Abstract]:In recent years, with the rapid development of information technology, with the emergence of a variety of intelligent terminals, applications and rapid growth of broadband users, network traffic is explosive growth, people's demand for high-quality content resources is also increasing day by day. Put a lot of pressure on the network infrastructure. Content delivery Network (CDN) is a technology that provides reliable and efficient content delivery. However, the traditional CDN can not obtain global network information and rely on DNS for request redirection, which leads to the lack of dynamic control over path and server selection. Software defined Network (SDN) is a new type of network technology, which separates the control surface from the data surface and provides flexible centralized dynamic management. The application of SDN technology in CDN network can monitor the network state information in real time and make the decision in time. At the same time, it can redirect the user's request in a finer time and fine granularity. Firstly, a CDN network architecture based on intelligent SDN is proposed. Intelligent SDN can effectively solve the problems such as increasing demand for SDN controller, increasing intelligent decision requirement and information sharing and interaction among multi-domain SDN controllers by introducing intelligent SDN center. This paper presents a CDN user request allocation algorithm based on Model Predictive Control (MPC). According to two core parameters related to user QoS: response time and bandwidth satisfaction, the candidate server and path selection are jointly optimized. Simulation results show that the performance of the algorithm is improved in response time and bandwidth satisfaction. It is also found that different priorities of response time and bandwidth adequacy can be reflected by adjusting the weight parameters to meet the needs of different network applications. In order to further optimize the performance, a CDN user request allocation algorithm combining neural network and MPC is proposed. The algorithm firstly scores different paths based on neural network and assigns user requests according to path score. Simulation results show that the performance of the algorithm is further improved because of the nonlinear mapping ability and self-learning ability of the neural network.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.0

【参考文献】