Design And Simulated Implementation Of A SDN-based Traffic Prediction And Scheduling Mechanism

As a new network architecture,Software-Defined Networking(SDN)decouples data plane from control plane and solves the complexity of traditional network configuration.It enables flexible network traffic control and makes network services more efficient and flexible.With the rapid development of SDN,the centralized control based load balancing problem becomes a hot research topic.It is crucial to design a reasonable traffic scheduling mechanism based on real-time network status and traffic characteristics.This thesis proposes a SDN-based traffic prediction and scheduling mechanism based on SDN,taking the advantage of SDN's centralized control to monitor network traffic information,and then using LSTM neural network models to predict network traffic in the next period.Traffic prediction is closely integrated with traffic scheduling to improve link load balancing and avoid network congestion.The main contents of this thesis are summarized as follows:Firstly,a general framework of traffic prediction and scheduling mechanism for SDN is proposed,mainly including traffic monitoring,traffic prediction and traffic scheduling.The traffic monitoring module gathers traffic statistics and port statistics periodically.Traffic information matrix from the source node to the destination node can be obtained from traffic statistics,and it is stored in the historical traffic information file to provide data basis for the traffic prediction mechanism.The port statistics information can be used to calculate the port flow rate and port residual bandwidth.The traffic prediction module implements a traffic prediction mechanism that treats source-to-destination node traffic data as a time series.It also performs data normalization processing,data set division and data shaping operations.Then this module constructs and trains LSTM neural network models to predict network traffic,then evaluates its accuracy.The traffic scheduling module puts forward a scheduling mechanism based on traffic prediction,including shortest paths acquisition,link remaining bandwidth calculation,reserved bandwidth calculation,predicted traffic information acquisition,optimal path selection,and reserved bandwidth.In the traffic scheduling process,the current network status,remaining path bandwidth,and reserved bandwidth are used as the conditions to select the best path.After the path is selected,the bandwidth is reserved according to the effective prediction value in the next period of the source-destination node,and the reserved bandwidth can be updates periodically.The simulation of the SDN-based traffic prediction and scheduling mechanism is developed on Ryu controller and Mininet platform.Using the TensorFlow+Keras deep learning library and the Statsmodels statistical modeling module to construct the LSTM model and the ARIMA model respectively.The Abilence network real history traffic data is used to evaluate the proposed traffic prediction mechanism.The simulation implements a scheduling mechanism based on traffic prediction.In order to verify the effectiveness of the scheduling mechanism proposed in this thesis,the simulation experiment is conducted based on the Abilence network topology.Experimental results show that the proposed traffic prediction algorithm has a lower mean absolute percentage errors of 9.62%.It presents a higher prediction accuracy compared with the Benchmark algorithm.The standard deviation of link utilization ratio of the traffic scheduling mechanism designed is one-third of the comparison algorithm.Results show that the proposed mechanism has a good load balance effect.