Research On Prediction-Based Resource Allocation Algorithm In Elastic Optical Network

With the rapid growth of network traffic,optical network technology is also constantly developing.With the emergence of Elastic Optical Networks(EON)and Software Defined Optical Networks(SDON),services can be served in a more efficient and flexible way.However,the resource allocation methods used now still lead to low resource utilization.To cope with the ever-increasing traffic,it is necessary to further optimize the resource allocation algorithm and reduce resource waste.Focusing on the resource allocation problem of time-varying services and asymmetric services in elastic optical networks,this dissertation studies traffic prediction to obtain more information of traffic and optimizes the resource allocation method based on this.The main work and innovative contributions are listed as follows:Firstly,the characteristics of network traffic are studied and a network traffic prediction model based Long Short-Term Memory(LSTM)neural network is proposed.The current network traffic presents complex nonlinearity,and LSTM network is suitable to solve such prediction problems because it can perform characteristic learning and extraction automatically.The simulation results show that the proposed model has better prediction accuracy than the traditional linear model.Through the prediction,the time-varying trend and asymmetry of services can be obtained in advance.Secondly,an Integer Linear Programming(ILP)model for time-varying services is proposed to minimize the number of connection reallocations of every time slot,which can decrease the network configuration costs caused by dynamic on-demand allocation.The simulation results show that the model can provide the optimization strategy.Furthermore,based on the prediction information,a resource allocation algorithm with selective resource reservation for time-varying services is proposed.In this algorithm,resource reservation is performed only for some part of services by comparing the predicted bandwidth demand and current bandwidth demand.Compared with the algorithm reserving resource for all services,the proposed algorithm can reach a compromise between the network blocking rate and the number of connections reallocated.Lastly,a holding-time-aware asymmetric spectrum allocation algorithm is proposed for the asymmetry of services' bidirectional bandwidth.The algorithm is designed for virtual optical networks(VON),because the services in VON have strong asymmetry,which leads that spectrum resources on the lightly loaded direction are redundant.The redundant resources in VON can be reused by other services to reduce spectrum resources' waste,and the simulation results show that the proposed algorithm can effectively improve the utilization of network resources.