基于SDN的多路径负载均衡算法及流表分配优化算法研究

发布时间：2018-02-13 08:47

本文关键词： SDN 数据中心网络多路径负载均衡流表　出处：《安徽大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着计算机网络的发展、大数据和云计算的出现,网络中数据流量爆炸式增长,软件定义网络(Software Defined Network,SDN)和数据中心网络成为了未来网络发展的主流方向之一。SDN定义了一种新型网络架构,它的核心思想是将网络的控制层和数据层进行分离,分别使用控制器和网络传输设备实现。与传统IP网络不同的是,SDN网络控制功能全部由控制器实现,网络传输设备仅用于数据的高速转发。数据中心网络解决了网络流量剧增带来的难题,通过将网络数据信息集中存储,实现了以信息为中心的网络。对SDN网络与数据中心网络的研究,已经得到越来越多研究者的关注。在数据中心网络中,为了提高网络带宽和吞吐量,多采用具有多路径特性的层次型网络拓扑结构,如胖树拓扑结构。然而传统路由算法对多路径的支持非常有限。针对此问题,在数据中心网络中引入SDN,通过对控制层的编程实现路由计算和网络集中控制,能够充分利用网络资源,实现多路径负载均衡。在SDN网络中,控制器通过标准的OpenFlow协议集中管理网络设备的转发行为。控制器制定路由策略后,向OpenFlow交换机下发流表,数据流使用匹配的流表项进行转发。交换机流表基于三态内容寻址存储器(Ternary Content Addressable Memory,TCAM)实现,只有大约1k-2k条流表项,流表资源有限。当数据流没有匹配的流表项时,会向SDN控制器发送packet-in消息请求分配流表资源。控制器每秒能够处理的packet-in消息大约只有30k条,计算资源有限。因此,在分配流表项时,需要均衡使用流表资源和计算资源,优化流表分配方案。论文的主要内容如下:(1)针对胖树数据中心网络中传统路由算法对多路径支持有限的问题,本文提出了基于路由评估的多路径负载均衡算法F-MPLB(Flow-multi path load balancing)。该算法利用SDN网络集中控制的特点,获取多路径实时状态信息,通过当前交换机节点和链路负载情况,评估多路径瓶颈带宽,以此选择最符合数据流传输需求的路径。实验结果表明,F-MPLB算法无论在降低网络传播时延还是在提高网络吞吐量等方面都优于传统路由算法,能够实现胖树数据中心网络的多路径负载均衡。(2)针对SDN网络中交换机流表资源和控制器计算资源有限的问题,本文在SDN网络中提出了基于资源偏好的流表分配优化算法RC-FDO(Resources consideration-flowtable distribution optimization),该算法通过引入流表资源代价、计算资源代价和资源偏好度,结合当前网络运行状态和数据流特性,研究如何设定的合理的流表项idle_timeout值。实验结果表明,RC-FDO算法能够有效提高流表资源利用率和数据流请求接受率,实现更高的网络平均吞吐量。
[Abstract]:With the development of computer network and the emergence of big data and cloud computing, the data flow in the network explodes. Software Defined Network (SDN) and data Center Network (SDN) become one of the mainstream directions of future network development. SDN defines a new network architecture, whose core idea is to separate the control layer and data layer of the network. Different from the traditional IP network, the SDN network control function is realized by the controller. The network transmission equipment is only used for high speed data transmission. The data center network solves the problem caused by the sharp increase of network traffic, by centralizing the storage of network data information, More and more researchers have paid attention to the research of SDN network and data center network. In data center network, in order to improve the network bandwidth and throughput, Hierarchical network topology with multipath characteristics, such as fat-tree topology, is often adopted. However, the traditional routing algorithm has very limited support for multipath. By introducing SDN into data center network, routing calculation and centralized network control can be realized by programming control layer, which can make full use of network resources and realize multipath load balancing. The controller centrally manages the forwarding behavior of network devices through the standard OpenFlow protocol. The data flow is forwarded using matching stream table entries. The switch flow table is based on the ternary Content Addressable memory (TCAM). There are only about 1k-2k stream table entries, and the flow table resources are limited. When the data flow does not match the stream table entries, Packet-in messages are sent to the SDN controller to request the allocation of flow table resources. The controller can process only about 30 k of packet-in messages per second, with limited computing resources. Therefore, when allocating flow table items, a balanced use of stream table resources and computational resources is required. The main contents of this paper are as follows: 1) aiming at the problem that the traditional routing algorithm has limited support for multipath in the fat tree data center network, This paper presents a multipath load balancing algorithm based on routing evaluation, F-MPLBN Flow-multi path load balancing.Using the characteristics of centralized control of SDN network, the algorithm acquires real-time state information of multipath, and obtains the load of current switch nodes and links. The experimental results show that the F-MPLB algorithm is superior to the traditional routing algorithm in reducing network propagation delay and improving network throughput. To solve the problem of limited resource of switch flow table and controller in SDN network, we can realize multipath load balancing of fat tree data center network. In this paper, RC-FDO(Resources consideration-flowtable distribution optimization algorithm based on resource preference is proposed in SDN network. By introducing flow table resource cost, resource cost and resource preference degree are calculated, combined with current network running state and data flow characteristics. The experimental results show that the RC-FDO algorithm can effectively improve the resource utilization of the stream table and the request acceptance rate of the data flow, and achieve higher average throughput of the network.
【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.02

【相似文献】