Research On Controller Capacity And Placement Problem In Software Defined Network

Software Defined Networking is a new emerging network architecture that decouples the control plane from data plane in traditional network based on software realization. The data plane is responsible for data transmission with specified hardware, while the control plane determines the flow routing via a centralized way. The decoupling between control plane and data plane can benefit the transparency of network facilities towards upper-level applications. Under the centralized management, the network resources can be better utilized.With the rapid development of SDN, it also confronts many challenges, in particular for the control plane. First of all, there should be a general system model to evaluate the network performance, especially for the controller capacity which is defined as the maximum number of switches the controller can manage, and how to deal with the scene when switch number goes beyond the controller capacity. Secondly, the decoupling of control plane and data plane brings a new problem named the controller placement problem, which is defined as where the controller should be placed to satisfy the optimal network performance. Lastly, in the scene of multiple controllers, the static controller placement strategy cannot adapt to dynamic flow variations, therefore a dynamic adaptive network management scheme is required.For the above issues, the thesis proposes a controller capacity evaluation method based on queuing theory, a SDN controller placement scheme under in-band control paradigm of controller, and an adaptive controller management scheme based on the flow dynamics under the scene of multiple controllers.To evaluate the controller capacity, we analyze the flow processing procedure in SDN, and model the flow set-up requests from the switches to controller as a batch arrival process. Furthermore, we analyze the controller performance with queuing theory, and derive the expression of average flow service time in the controller. Under the circumstance of a limited flow set-up time (such as the QoS demand of network flows, the limited buffer size of switches and so on), the number of switches is then determined which provides a method to evaluate the controller capacity.We propose a node-weight considered controller placement scheme to optimize the network performance. In real network, the node weights are different from each other. The bigger the node weight, the more flow requests it will generate. To minimize the cost of new flow set-up requests from switches to controller, the nodes with bigger weight should be considered more when deploying the controllers. From this view, we evaluate the node weight with node degree; meanwhile, we define the candidate controller set to decrease the complexity when choosing optimal controller. And the deploying instance indicates that the node-weight considered controller placement scheme can minimize the flow set-up cost, therefore achieving optimal network performance.The controller placement problem is the work during network initialization, the positions of controllers are constant and static from then on. However the static controller placement scheme cannot adapt to the flow dynamics. We propose an adaptive controller management procedure based on the flow dynamics. Due to the different flow variations, the network can take corresponding decisions. Furthermore, we propose a dynamic switch migration algorithm to adapt to the flow dynamics and realize controller load balance in multiple SDN domains; meanwhile the scheme can guarantee the QoS demand of migrated switch. A simulation platform is built to verify the proposed scheme, and the simulation results demonstrate that the proposed algorithm can achieve the load balancing of controllers, therefore realizing the efficient utilization of control plane resources in SDN.