基于层次聚类算法的SDN分布式控制器的网络视图优化

发布时间：2018-06-17 21:27

本文选题：软件定义网络 + 分布式控制器　；参考：《吉林大学》2017年硕士论文

【摘要】：自从2008年以来,SDN在提出之后短短几年时间就取得了飞速的发展,SDN的最大特点就是控制平面和转发平面相互解耦,通过一些特殊的手段把网络中的控制平面从底层网络中解耦出来,通过可编程,可操控的控制平面取代了传统的较为封闭的嵌入式控制平面,提取出来的控制平面将会集中到一个集中式控制器中,这样一来,整个网络就会由这个集中式控制器来负责,这样就能够达成了动态管理以及灵活调度网络资源的目标。随着社会发展,网络规模不可避免的增大,一个单一的集中式控制器很容易出现过载或者故障。对于大规模的网络来说,即使是一个非常强大的控制器,也会因为缺乏必要的处理器和内存容量从而很难保持完整的网络状态,也就无法对所有的网络事件做出反应。如果发生这种事件,整个网络的服务性能就会不可避免的降低,严重情况下甚至会导致网络瘫痪,这就需要一个分布式集群,这个集群有许多个SDN控制器构成,这样就能够解决单一控制器的性能问题。现在,为了解决单一控制器所面临的各种各样的问题,许多研究机构和企业都在单一的集中式控制器的基础之上提出新的架构,主流的的架构主要有两种:一种是以ONOS控制器为代表的拥有一个集中式数据库的分布式控制器,另外一种是以ONIX控制器为代表的完全分布式控制器,以上两类控制器都可以避免单一控制器所面临的性能不足的问题,同时也有着各自独有的优点和缺点。本文研究了以上两种分布式控制器,并且在分布式控制器架构的基础上设计了资源开销模型,主要包含存储开销和通信开销这两种比较重要的开销。并且在这个资源开销模型的基础之上提出了自己的优化算法。本文的优化算法主要是在两种较为主流的控制器架构中寻找到一种平衡,进一步的在存储开销和通信开销之间寻找平衡,对控制器的网络视图进行从新分配,从而更加合理的使用网络中的资源,最终达到减少总的资源开销的目的。本文还通过多组的实验对比来对算法的结果进行了论证,并且进一步探讨算法的使用环境,最终本文对未来的研究方向提出了许多理论上的规划。
[Abstract]:Since its inception in 2008, SDN has made rapid development in a few years. The biggest characteristic of SDN is that the control plane and the forwarding plane are decoupled. The control plane in the network is decoupled from the bottom network by some special means. The programmable and controllable control plane replaces the traditional closed embedded control plane. The extracted control plane will be centralized into a centralized controller, so that the whole network will be responsible for the centralized controller, which can achieve the goal of dynamic management and flexible scheduling of network resources. With the development of society, the scale of network increases inevitably, and a single centralized controller is prone to overload or failure. For large-scale networks, even a very powerful controller will be unable to respond to all network events because of the lack of necessary processors and memory capacity to maintain a complete network state. If such an event occurs, the service performance of the entire network will inevitably decline, and in serious cases even lead to network paralysis, which requires a distributed cluster, which has many SDN controllers. In this way, the performance of a single controller can be solved. Now, in order to solve the various problems faced by a single controller, many research institutions and enterprises have proposed a new architecture based on a single centralized controller. There are two main architectures: one is a distributed controller with a centralized database represented by an ONOS controller, the other is a fully distributed controller represented by an ONIX controller. These two kinds of controllers can avoid the problem of insufficient performance faced by single controller, and also have their own advantages and disadvantages. This paper studies the above two distributed controllers, and designs a resource overhead model based on the distributed controller architecture, which mainly includes storage overhead and communication overhead. On the basis of the resource overhead model, this paper proposes its own optimization algorithm. The optimization algorithm in this paper is mainly to find a balance between the storage overhead and the communication overhead in the two main controller architectures, and to redistribute the network view of the controller. Thus more rational use of resources in the network, and ultimately achieve the purpose of reducing the total cost of resources. This paper also demonstrates the results of the algorithm through the comparison of many experiments, and further discusses the use environment of the algorithm. Finally, this paper puts forward many theoretical plans for the future research direction.
【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】：TP311.13;TP273

【相似文献】