| Software-Defined Networking,as a new type of network architecture,has decoupled control plane from data plane,and solves the problems of the rigid structure,difficult management and complex configuration existed on the traditional TCP/IP network.Therefore,SDN is closely watched by the academic field and industry field.As the network scale expands steadily and traffic increases rapidly,the single controller is limited by scalability and reliability when managing the network.Thus,it is proposed that multi-controller multi-domain scheme can effectively improve the elasticity of SDN control plane.In terms of multi-controller network,the processing performances of controllers may be quite different due to the dynamic change of network traffic,unreasonable multi-controller deployment and the static connection between switch and controller.It is easily to generate overloaded or underloaded controller,which degrades the performance of SDN network.Therefore,how to balance multi-controller load in large-scale SDN network has become a hot issue in recent years.The existing researchers mainly focus on three different levels to achieve the load balance of multi-controller: balanced planning for SDN multi-domain in the initial construction phase of the network;adjusting the deployment of multi-controller in the process of constructing the network;migrating switch after the network has been constructed successfully.However,the above schemes still have the following problems to be improved:(1)in the multi-domain planning level,there are insufficient researches on the bidirectional mapping between switches and controllers;(2)in the controller deployment level,the traditional deployment scheme is difficult to simultaneously consider the reliability and load balancing performance of the control node;(3)in the switch migration level,the unreasonable selection of migration objects and the complicated design of migration mechanism will lead to the low migration efficiency.Relying on the national high-tech research and development plan(863)project “software-defined network architecture and key technologies”,we set the multi-controller load balance as the research target and analyze and research the load balance from the perspective of the multi-domain planning,controller deployment and switch migration to improve the load balance performance of multi-controller.The research work and main contributions of this paper are as follows:1.We propose a multi-domain planning method based on bidirectional matching to solve the unreasonable multi-domain partition in the initial construction phase of the network.Through collecting hop counts,delay and flow information in the network periodically,we build match lists of switch and controller respectively.According to the principle of optimal queuing,it selects switches and controllers from two match lists for implementing bidirectional matching,and optimizes the relationship of matching with the help of the simulated annealing algorithm,which achieves dynamic deployment for multi-controller in the distributed network.The feasibility of the method is demonstrated through mathematical derivation and analysis.The simulations prove that this method can realize the stable connection between the switch and the controller,and the SDN multi-domain is planned in a balanced way.The multi-controller load balancing rate has been improved 21.4% at least.2.We propose a multi-controller balanced deployment method based on reliability evaluation in SDN to solve the problem of combining reliability and load balance of the control plane in the process of constructing the network.Firstly,Node reliability is evaluated by weighting node efficiency and path quality,optimizing the controller location with balancing factor.Then based on improved K-Center clustering,we introduce the redundant functions and complete the switch allocation according to node attractiveness degree and controller load balancing rate,achieving a reliable and balanced deployment of multi-controller The simulations show that,compared with the existing methods,the control plane elasticity is obviously enhanced,the required number of controllers has been reduced 22.1% averagely,and the load balancing performance of the controller is improved greatly.3.We propose a dynamic switch migration method based on distributed policy to solve the rigid selection of emigration controller and migration conflict of switch after the network has been built.The method is divided into three phases.Firstly,through collecting network information periodically,distributed migration decision domains are structured according to controller load conditions.Then the switches are migrated according to selecting probability.By balancing three costs included data collection,switch migration and state synchronization,the target controllers are determined.Finally,this mechanism sets the migration countdown to complete the switch migration and controller role conversion.The simulation results show that this method can reduce communication cost effectively,shorten the average building time of the flow by 0.14 s and improve the resource utilization of the controller by 21.7%. |
PDF Full Text Request