Research On Mechanism Of Fast Failover For Multi-domain Software-Defined Networking

With the rapid development of Internet,the scale of network users is expanding rapidly,and the static structure and management mode of traditional network are becoming more and more inappropriate to the dynamic development trend of various applications.As a new type of network architecture,software-defined networking is easy to manage and is programmable,which can solve the problems of complexity of network equipment and difficulty in configuration of traditional network.The typical single controller software definition network can realize the network construction on the basis of inheriting the SDN characteristics.However,the single controller network has some inherent defects that could not be overcome: one is the problem of single point failure,and the other one is the problem of network scalability.Therefore,the industry has proposed a control system under a multiple controller architecture,which through two ways to solve the controller single point of failure problem.Firstly,in the event of failure,locate the fault controller through the corresponding detection mechanism.Then,redistribute the switches under the fault controller to the other controllers.However,the above two methods have the following problems: first,the existing fault detection methods could not quickly locate the fault and the detection point is not reliable;second,the existing switch redistribution method is too extensive.Therefore,the specific content of this thesis is as follows:Firstly,in the aspect of fault detection,a fault detection model based on gray prediction using an adaptive fault detection algorithm is proposed to overcome the problem of the current fault detection point is single and the adopted fixed timeout interval.In the aspect of detection mode,the ring detection method is adopted to ensure the reliability of multiple fault detection points.In the aspect of detection algorithm,for the dynamic change of the network,an adaptive prediction method is adopted to record the previous messages and predict the next message,then,threshold values are set accordingly.Simulation results show that the proposed algorithm can achieve fast fault location and ensure the reliability of the detection point.Secondly,in the case of switches redistribution strategy,a redistribution strategy based on loss degree is proposed,which consists of three parts.Firstly,when the controller fails,the corresponding model is established based on the load of the controller,the shortest distance between the switch and the controller,and the loss degree of the switch during the redistribution process.Secondly,the objective functions of the model are calculated according to different mapping methods,and select the optimal result.Finally,the redistribution mapping matrix of switches and controller is obtained,which is the result of switches redistribution.Simulation results show that the strategy can make the switches in the whole fault domain obtain the flow list quickly,and reduce the probability of the connection failure between the switches and the controller.