Research On The Reliability Of Software Defined Network

Software defined network (SDN) is a new network architecture, whose key idea is to separate control layer from forwarding layer. The controller in the control layer utilizes centralized management of forwarding layer, which can obtain the topology information of the whole network. According to the topology information of the whole network, controller can provide applications, such as load balancing, path optimization, flow distribution and so on.When the network scale is small, a controller can be satisifing to centralized management of the whole network, accessing controller topology information simple-ly and directly.But when the network scale is large, relying on a single controller has been unable to meet the needs of management of the whole network, and multi-controller joint management of the whole network appears. In the mode of multi-controller management, each controller only can manage a subnetwork of the whole, collect the subnet information and be responsible for the subnet management, and the sub network resource allocation performance. Controller shares its subnetwork topology information among each other, in order to obtain the complete topology information of the whole network. However, if one cannot obtain the correct subnetwork information due to link failures in the network, then, the other controller also cannot get the whole network topology information completely, resulting in a variety of applications depending on the controller making mistakes. This paper put forward two proposals to solve this problem. In these two schemes, each controller independently detectes link state, link failures and locate link failures appeared in the network. The contribution of this thesis is as follows:(1) According to the existing scheme,controller is weak to locate multi-link faults in SDN.This paper proposed a polycyclic generatin algorithm based on genetic algorithm, the algorithm can generate cycles to cover all the links in the network. Locating multi-link faults relied on the cycles in the network. At the same time, the performance of the generated polycyclic algorithm is optimized, which makes the number of repeated links contained as little as possiblely. The experimental results show that the scheme is superior in the ability to detect multiple link failures than existing schemes;(2) In order to overcome the problem that the existing scheme is difficult to adapt to the situation of network link additions and deletions, this paper proposed link fault localization scheme based on tree decompositon. This schema improves the existing graph segmentation algorithm, the tree decomposition of losses graph partitioning algorithm being able to partition the network into multiple independent sub networks is proposed based on the algorithm, compared with the traditional graph segmentation algorithm to generate the subnetwork has unique characteristics:first, the nodes in each subgraph with connectivity and node degree is greater than or equal to2; Secondly,both independence and connectivity exists among the generated sub graphs. The network topology is divided into multiple sub networks, reducing the effect of a single link in the net-work changes bring to the whole network. In order to adapt to the pretreatment of the single controller network topologyscheme, this schema further puts forward the maximum degree of breadth first tree decomposition multi cycles generation algorithm, this algorithm makes the ability of polycyclic single controller link fault detection and location in the indepent subnet.Experimental results show that the number of indicators such as static flows in the schema needed to be sent to the flow table and detection information needed to be sent to detect link faults is superior than existing schemes.