Research On Multi-Controller Optimization Deployment Strategy In SDN Environment

On the one hand,the software-defined network has changed the traditional network control and forwarding tightly-coupled working modes through numerical control separation of ideas,which has improved the flexibility and openness of network management.On the other hand,the northbound interface based on the controller itself enables the network to have a strong programmability,which enables developers to flexibly define network applications through software,to meet the personalized needs of users on the network,and to bring intelligence to network management.Endless possibilities.Specifically,a Software Defined Network is not a specific network protocol but a network architecture.A network that generally has these two characteristics can be referred to as a broad Software Defined Network.In large-scale networks,the network performance is degraded due to the limitation of controller performance in Software Defined Network.Therefore,the study of control plane scalability is necessary.The research shows that deploying multiple controllers on the control plane to relieve the overloaded single controller is an effective method to solve the problem of control plane scalability.This article focuses on the deployment strategy of multi-controllers in Software Defined Network.The main contributions are reflected in the following three aspects:(1)Study the factors that affect the performance of the controller,and propose a controller optimization algorithm CNO based on the controller's processing capability index.By analyzing the characteristics of the arrival of the new flow to the controller and the controller's handling of the new flow,the mathematical theory of the process is mathematically modeled using the related theories of queuing theory and the mathematical expression of the model is obtained.By setting related parameters of the model,the model is finally solved mathematically,and a reasonable number of controller deployments in the SDN network is determined.In the actual application process,the required number of controllers can be directly calculated through the relevant parameters of the control plane,and the process of processing the new flow by the control plane is modeled by the queuing model,and the obtained result is closer to the actually measured stay experiment.(2)Study the SDN network topology structure,improve the extended random block(GSB)model,and add the load balance factor in the GSB model,and propose a balanced controller deployment algorithm CDGSB with propagation delay as the main indicator.According to the characteristics of the community structure in the network topology,the extended random block model is used to fit the network generation process,and the model parameters are estimated using the expectation maximization method in statistics,thereby utilizing the internal network without any prior information.The structure divides the network into several sub-domains and adjusts the divided sub-area results twice according to the load balancing indicator to adapt to the control scale of the SDN controller.When determining the deployment position of the controller in the sub-domain,the shortest propagation delay of the controller is used as the objective function,and the deployment algorithm is designed with the load balancing index as the constraint condition.The algorithm can meet both the internal structure of the network and the load balance and the network.Delay required controller deployment strategy.(3)To study the reliability of SDN network links,improve the Trajin algorithm for finding biconnected components in the graph structure,and propose a multiple controller deployment algorithm Reliability-Trajin with reliability as the main indicator.Using the characteristics of joint points in graph theory,based on the reliability index,the concept of the dual connected components in graph theory is used to deal with the problem of sub-domain division of small and medium-sized networks.Find the top of the network topology graph and use a stack to store the traversed edges during the DFS tree traversal so that the edges can be taken directly from the stack after finding the top of the cut,that is,the edges of the double connected components.Then use the minimum average hop count from the switch in the domain to the controller as the objective function to avoid the cut point position and finally determine the actual location of the controller deployment.By separating the method of sub-domain division and controller location deployment,the complexity of the algorithm can be reduced,which not only makes the algorithm structure clear,but also can achieve the controller deployment strategy that meets the requirements of reliability and load balancing simultaneously,and is more in line with the reliability-sensitive SDN.