基于SDN的数据中心网络流量工程研究

发布时间：2018-03-01 12:16

本文关键词： SDN 数据中心流量工程负载均衡胖树拓扑 LLDP　出处：《杭州电子科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：数据中心作为重要的信息化基础设施,在云计算飞速发展的过程中一直处于核心地位。数据中心网络流量与互联网流量有很大不同,新兴的业务致使不同的虚拟机之间产生大量的数据中心内部流量。同时,数据中心规模的爆炸式增长让传统数据中心网络力不从心,不充分的资源分配带来了链路拥塞和端到端时延的增大,严重制约着数据中心性能和服务质量进一步提高,一种新的网络管理模式势在必行。基于SDN的数据中心网络有着传统数据中心网络无法比拟的优势。SDN控制器基于全局网络视图,结合当前网络资源的实时性能和状态为数据流量做出最优的路由判决,从而达到更好的负载均衡效果。本文从以下五个方面进行研究:(1)概述了数据中心网络的研究背景和意义,对SDN技术在多种网络场景下的应用现状进行了归纳,着重研究了SDN在数据中心网络流量工程中的应用现状。(2)阐述了SDN的定义和架构,包括OpenFlow协议、控制器、交换机以及控制器与交换机之间的安全通道,对各Open Flow协议版本进行了比较,然后着重介绍了本文中使用的Ryu控制器及其开发流程。(3)建立了基于数据中心Fat-Tree网络的ECMP(Equal Cost Multi Path)数学模型,然后在此基础上提出了基于端到端链路利用率的ECMP模拟退火算法(SA-ECMP),并使用该算法为等价路径计算合适的流量分配比例。(4)介绍了本文中Ryu控制器的总体思想,并分模块阐述了系统的实现原理,包括基于LLDP的拓扑发现模块、链路状态获取模块、流表下发模块和拓扑视图模块。(5)利用Mininet网络仿真平台,实现了对数据中心网络的模拟,并对本文提出的SA-ECMP算法进行了仿真验证。该算法相比于文献中提到的其他算法,提高了数据中心网络中的链路带宽利用率,降低了端到端传输时延,收到了较好的负载均衡效果。
[Abstract]:As an important information infrastructure, data center has always been in the core position in the rapid development of cloud computing. Data center network traffic is very different from Internet traffic. New businesses have created a lot of internal traffic between different virtual machines. Meanwhile, the explosion in the size of data centers has left traditional data center networks unable to cope. Inadequate resource allocation leads to the increase of link congestion and end-to-end delay, which seriously restricts the performance of data centers and the further improvement of quality of service. A new network management mode is imperative. The data center network based on SDN has unparalleled advantages over the traditional data center network. The controller is based on the global network view. Combined with the real-time performance and state of current network resources, the optimal routing decision is made for the data traffic, so as to achieve better load balancing effect. In this paper, the research background and significance of data center network are summarized from the following five aspects: 1. This paper sums up the application status of SDN technology in various network scenarios, and emphatically studies the application status of SDN in data center network traffic engineering.) the definition and architecture of SDN are expounded, including OpenFlow protocol, controller, etc. The Open Flow protocol versions are compared between the switch and the secure channel between the controller and the switch. Then, the paper introduces the Ryu controller and its development flow in this paper, and establishes the mathematical model of ECMP(Equal Cost Multi path based on the Fat-Tree network of data center. Then, an ECMP simulated annealing algorithm based on end-to-end link utilization is proposed, and the algorithm is used to calculate the appropriate traffic allocation ratio for equivalent paths. The general idea of the Ryu controller in this paper is introduced. The realization principle of the system is described in detail, including the topology discovery module based on LLDP, the link state acquisition module, the stream table sending module and the topology view module. (5) using the Mininet network simulation platform, the simulation of the data center network is realized. The proposed SA-ECMP algorithm is verified by simulation. Compared with other algorithms mentioned in the literature, the proposed algorithm improves the bandwidth utilization of the data center network and reduces the end-to-end transmission delay. Good load balancing effect has been obtained.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.0;TP308

【参考文献】