| The transmission of traffic in the traditional network relies on distributed routers through self-discipline control such as routing protocols and algorithms,and the data link and network layer have both control and forwarding ability.Its control capability is distributed,and each network forwarding device is integrated and autonomous.This makes them extremely complex,difficult to manage and to dynamically modify network configurations in response to failures,loads and updates.Software-Defined Networking(SDN)overcomes the limitations of traditional networks and has the advantages of flexible management and centralized control.However,with the rapid growth of network scale,in the distributed architecture,each controller manages the forwarding devices in the domain respectively,and the traffic of the forwarding devices in each domain is unstable,which causes unbalanced load distribution among them.Therefore,this dissertation studies the problem of how to dynamically balance the load between controllers,and the main contributions are as follows:(1)Prediction-based dual-weight switch migration scheme for SDN load balancing is proposed.The migration speed of the switch is affected by the current controller load.If switch migration is necessary,the earlier the migration is done,the faster the migration can be.This dissertation designs a prediction-based dual-weight switch migration scheme for SDN load balancing.The scheme predicts the occurrence time of the controller’s high load in advance to start the load balancing operation.At the same time,a triggering load information algorithm is proposed,which solves the periodic passive load information strategy between distributed controllers and reduces the extra processing and communication overhead of the controllers.To reduce the migration frequency of switches,this dissertation uses the historical load and future load to propose a dual-weight switch migration algorithm,and the effectiveness of the scheme is proved by experimental comparison.(2)An improved entropy SDN load balancing scheme based on ARIMA is proposed.The emergence of distributed architecture solves the single point of failure and scalability problems brought by centralized architecture.However,the lack of a flexible traffic balancing mechanism among multiple controllers can easily lead to uneven load distribution.For the distributed SDN control plane,an improved entropy SDN load balancing scheme based on the Autoregressive Integrated Moving Average model(ARIMA)is proposed.The scheme uses the ARIMA model to perform multi-step prediction based on the past traffic data and obtains the multi-step traffic prediction result.At the same time,a trigger load information synchronization strategy is proposed to reduce the extra overhead caused by information synchronization.The improved entropy-based switch selection algorithm and optimal controller selection algorithm proposed by the application scheme can solve the problem of frequent switch migration and reduce the number of load balancing operations.Finally,the comparative experiments show the advantages of this scheme. |
