Research On Multi-controller Placement Problem In Software Defined Network

Software Defined Network has been attracting a growing attention in the network research community because of its characteristic of decoupling control function from data forwarding devices. The control layer, which is indispensable in SDN, is a bridge between the network applications and data forwarding devices and its performance directly influences the performance of the whole SDN network. When deploying SDN to Wide Area Networks, it's hard to meet the demands of network services using only a single controller. It is because of the limited capacity and the single-point-of-failure of a single controller. As a result,the network performance of control layer is difficult to be guaranteed.The distributed controller placement strategy is an effective approach to solve this problem. And the controller placement problem under this strategy will directly influence the performance of control layer.The existing papers on the controller placement problem are almost all considering only one single factor, such as reliability, latency,resilience and so on. They have ignored the overall performance of the control layer mostly. This paper analogizes the controller placement problem considering both the latency and the reliability of the network.Firstly, in order to quantify the reliability of the control network, this paper defines the reliability evaluation index of control layer as the expected percentage of node inaccessibility based on the definition of network reliability. The index means that the expected percentage of nodes those cannot connect to the control network, which can intuitively reflect the effects of network element failure on the network performance.On this basis,this paper then proposes a high reliable controller placement model on the constraint of propagation latency and gets a better solution with the highest reliability and the minimum latency. At last, a heuristic algorithm based on simulated annealing is proposed to reduce the complexity of the model and to make it more suitable for large scale networks.The simulating tests of the model and the heuristic algorithm in the SNDlib topologies demonstrate the effectiveness of the model and algorithm. The results show that in most cases, the model and the algorithm proposed in this paper can effectively reduce the worst-case latency while ensuring the optimal reliability of the network. This has a profound significance for information transmission in Wide Area Networks.