文件名称：软件定义网络
文件大小：180KB
文件格式：PDF
更新时间：2017-06-04 15:27:49
sdn
Unlike traditional networks, where both control and for- warding planes are highly integrated on the same boxes, the Software Defined Network (SDN) architecture decouples control and forwarding planes. Such separation is realized by moving the network intelligence onto one or more external servers, called controllers, which make up a network-wide logically centralized control plane that oversees a set of dumb, and simply forwarding elements [1]. Generally speaking, the SDN control plane consists of three parts [2]: control platform, which handles state distribution; control applications, which are developed upon a program- matic interface that the control platform provides; and control network (or connectivity infrastructure), which is used for propagating events to packet-forwarding devices or between multiple controllers. Although there have been a range of previous works that focus on control platform and control applications, so far, design issues of the control network, which is also another essential part of the SDN control plane, have not been well investigated. Since network failures could cause disconnections between the control and data planes, and further disable some of the switches, it is of great importance to improve the reliability of SDN control networks. As a result, in this paper, we are inter- ested in finding the answers to the following questions: given a physical network and the failure probability of each network component (e.g. links and switches), how many controllers are needed and how to place them such that a pre-defined reliability objective is optimized.