基于SDN的数据中心网络多路径负载均衡的研究

发布时间：2018-03-13 00:33

本文选题：SDN　切入点：负载均衡　出处：《大连海事大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着近年来互联网飞速发展,网络流量呈现爆发式增长,数据中心作为其重要基础设施也面临着诸多挑战。软件定义网络(Software Defined Network,SDN)是对传统网络架构的一项重大变革,其核心理念是将数据层面与控制层面分离。本课题主要对基于SDN的数据中心网络多路径负载均衡策略进行了研究。针对数据中心网络流量集中的特点,利用SDN集中式控制和软件可编程的优势,通过多路径负载分担的方法提高网络链路利用率,减少拥塞。本文选择Ryu控制器作为研究平台,通过REST API在应用层自定义5个模块来实现多路径负载均衡策略,各模块分别为.全局拓扑感知模块,网络状态测量模块,k短路径计算模块,链路状态评估模块和拥塞处理与流表转发模块。Ryu控制器周期性地调用全局拓扑感知模块和网络状态测量模块,来获取网络全局拓扑、带宽、时延等信息,以实现动态监控。通过k短路径计算模块、链路状态评估模块、拥塞处理与流表转发模块来实现当前流量负载合理分配的功能。Ryu控制器能够通过周期性发送LLDP报文来获取全局网络拓扑信息,并且可以使用软件方法防止网络环路和实现MAC地址学习,利用OpenFlow协议可以实时监测网络带宽、时延等参数。本文选用Yen算法作为KSP算法计算出k条备选路径,并建立路径评估模型,评定各备选链路优先级,选择当前优先级最高的链路作为转发路径,并将对应的流表项下发到OpenFlow交换机。周期性监控链路拥塞情况,在当前转发路径的带宽利用率达到阈值时,选择除当前路径以外的最优路径实现重路由。本文利用Mininet仿真平台模拟数据中心网络胖树拓扑结构,并在Ryu控制器上实现本课题多路径负载均衡策略,使用Mininet自带的Iperf打流工具模拟流量发生,进行两组对比实验。实验结果表明:本文调度策略在提高网络链路利用率,降低时延,减少拥塞等方面有较好的效果。
[Abstract]:With the rapid development of the Internet in recent years, network traffic is exploding, and data center is facing many challenges as its important infrastructure. The core idea is to separate the data level from the control level. This paper mainly studies the multi-path load balancing strategy of the data center network based on SDN, aiming at the characteristics of the data center network traffic concentration. Taking advantage of the advantages of SDN centralized control and software programming, the multipath load sharing method is used to improve the network link utilization rate and reduce congestion. In this paper, the Ryu controller is chosen as the research platform. The multipath load balancing strategy is implemented by using REST API to customize five modules in the application layer. The modules are. Global topology perception module, network state measurement module and short path calculation module. The link state evaluation module and the congestion processing and flow table forwarding module .Ryu controller periodically call the global topology sensing module and the network state measurement module to obtain network global topology, bandwidth, delay and other information. In order to realize dynamic monitoring. Through k short path calculation module, link state evaluation module, Congestion processing and flow table forwarding module to realize the function of reasonable distribution of current traffic load. Ryu controller can obtain global network topology information by sending LLDP packets periodically. The software method can be used to prevent the network loop and realize MAC address learning, and the network bandwidth and delay can be monitored in real time by using the OpenFlow protocol. In this paper, the Yen algorithm is chosen as the KSP algorithm to calculate k alternative paths. A path evaluation model is established to evaluate the priority of each alternative link, and the link with the highest priority is selected as the forwarding path. The corresponding flow table items are sent to the OpenFlow switch, and the link congestion is monitored periodically. When the bandwidth utilization of the current forwarding path reaches the threshold, the optimal path other than the current path is chosen to realize the rerouting. This paper simulates the fat tree topology of the data center network by using the Mininet simulation platform. The multipath load balancing strategy is implemented on the Ryu controller, and the traffic generation is simulated by using the Iperf routing tool of Mininet. The experimental results show that the scheduling strategy in this paper can improve the network link utilization ratio. It has good effect in reducing delay and congestion.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP308;TP393.0

【相似文献】