SDN网络下基于BP神经网络算法的负载均衡研究

发布时间：2018-01-29 22:09

本文关键词： SDN网络 BP神经网络算法负载均衡　出处：《吉林大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,随着移动互联网、电子商务的不断发展,大数据和云计算成为新的发展趋势。网络流量压力越来越大的同时,流量模式逐渐变为高度动态化,按需提供带宽的能力代替了总传输容量增益成为创造收益的关键。服务提供者基于业务的多样化和建设成本考虑,对灵活优化的管控能力更加渴求;IT基础设施也要应需而变,以全新的方式支持快速灵活的商业运作。SDN(软件定义网络)就是在这种情况下应运而生。它是一种新型的网络架构,是对传统tcp/ip网络架构的一种伟大的革新与突破。顾名思义,SDN的思想就是通过软件定义和驱动的方式实现对网络的控制和管理。SDN架构分为应用层、控制层、数据层。数据层上的网络设备专注于数据包的转发从而提高转发速度;控制层由SDN控制器通过编程来集中管理调度,可以结合当前网络状态和用户需求来制定转发和调度策略;应用层上的开发人员通过SDN控制器获知了网络的全局视图,可以灵活自由的部署业务。然而由此也引出了另一个新的问题,如何在提供服务质量的同时,实现对资源的合理分配,实现在SDN网络中进行负载均衡。负载均衡也是近年来越来越热门的话题。面对越来越多的新型应用,越来越大的网络流量,网络承受的压力也越来越大。要保证服务的质量一方面需要提高数据处理能力和响应速度,另一方面需要进行负载均衡,实现网络资源合理分配。那么在SDN网络下以何种方式进行路由决策来平衡负载提高网络性能成为一个新的研究热点。本文针对当前网络饱受各种压力的现状及传统网络架构面临的瓶颈,通过对新兴网络架构SDN网络的研究,提出了在SDN下基于BP神经网络算法的负载均衡方案。SDN网络的一大特点是控制层和数据层分离,控制器能够获取整个网络的拓扑结构,同时能够获得交换机传来的实时网络状态信息,包含链路负载、时延等等,我们便可以利用这一点在控制层上制定某种路由策略对数据包进行转发,实现负载均衡。我们可以利用启发式算法来获得某种场景下的一个近似最优解,但是由于它所需的时间成本很高,所以我们采取它的结果作为一个训练工具,一个中间产物。BP算法属于人工神经网络中的一种算法,它的特点是具有自学习能力和泛化能力,能对输入-输出关系进行模型训练,当训练完成后对输入具有快速预测输出的能力。所以本文的方案就是将实时网络状态和Qo S请求作为BP算法的输入,将启发式算法求得的最佳路由作为BP算法的输出,将此模型进行训练,训练后的模型(路由决策)存放在SDN控制器中,当有新的请求时可以快速预测出路径,从而使得SDN网络下负载均衡达到更好的效果。
[Abstract]:In recent years, with the continuous development of the mobile Internet and e-commerce, big data and cloud computing become a new development trend. The ability to provide bandwidth on demand replaces the total transmission capacity gain as the key to generating revenue. Service providers are more eager for flexible and optimized control capabilities based on business diversification and construction cost considerations; The IT infrastructure has to change as well, in a completely new way to support fast and flexible business operations. SDN (Software defined Network) emerged as the times require. It is a new network architecture. It is a great innovation and breakthrough to the traditional tcp/ip network architecture. The idea of SDN is to realize the control and management of network by software definition and drive. SDN architecture is divided into application layer and control layer. Data layer. The network devices on the data layer focus on the forwarding of data packets to improve the speed of forwarding; The control layer is managed centrally by the SDN controller through programming, and the forwarding and scheduling policies can be formulated according to the current network status and user requirements. Developers in the application layer know the global view of the network through the SDN controller, and can deploy the service flexibly and freely. However, this also leads to another new problem, how to provide the quality of service at the same time. Realize the rational allocation of resources and realize load balance in SDN network. Load balancing is also a hot topic in recent years. Facing more and more new applications, more and more network traffic. The network is under increasing pressure. To ensure the quality of services, on the one hand, we need to improve the data processing capacity and response speed, on the other hand, we need to balance the load. How to make routing decision in SDN network to balance load and improve network performance has become a new research hotspot. This paper aims at the current situation of network under various pressures. And the bottleneck of traditional network architecture. Through the research on the new network architecture, SDN network, a load balancing scheme based on BP neural network algorithm under SDN is proposed. One of the characteristics of the network is the separation of control layer and data layer. The controller can obtain the topology of the whole network and the real-time network state information from the switch, including link load, delay and so on. We can use this point in the control layer to formulate a certain routing policy to forward packets to achieve load balancing, we can use heuristic algorithm to obtain an approximate optimal solution in a certain scenario. But because the time cost is very high, so we take its result as a training tool, an intermediate product. BP algorithm belongs to an artificial neural network algorithm. It has the ability of self-learning and generalization, and can train the model of input-output relationship. When the training is completed, it has the ability to predict the input and output quickly. Therefore, the scheme of this paper is to use real-time network state and QoS request as the input of BP algorithm. The best route obtained by the heuristic algorithm is taken as the output of BP algorithm. The model is trained and the trained model (routing decision) is stored in the SDN controller. When there are new requests, the path can be predicted quickly, so that load balancing in SDN network can achieve better results.
【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】：TP183;TP393.0

【相似文献】