Routing And Spectrum Assignment Algorithms With Prediction In Elastic Optical Networks

With the increasing emergence of new applications on the Internet,all kind of services with great variation in bandwidth requirements also arise.The traditional optical networks based on wavelength division multiplexing(WDM)are more difficult to simultaneously meet the needs of both users and network operators,which results from its fixed and coarse allocation granularity in unit of a wavelength.Elastic optical networks(EONs)based on orthogonal frequency division multiplexing(OFDM),which allocate spectrum resources more efficiently and flexibly by a finer granularity partition for spectrum according to the users' bandwidth requirements,have been considered as a promising candidate for future high-speed optical networks.Routing and spectrum assignment(RSA)is one of the key issues in EONs and also the focus of this thesis.Firstly,the thesis summarizes the research background and significance,and outlines the research status of optical networks,RSA and applications of prediction technology in the network field.Secondly,the basis of EONs and RSA is introduced,and the existing optical network traffic models are described.The basic principles of the prediction method adopted in this thesis are explained in detail after summarizing the existing prediction models.To reduce the traffic blocking probability and increase the network resource utilization,a new RSA algorithm based on Traffic Prediction and Periodic Rerouting(TPPR)is proposed.The algorithm predicts traffic variations with time on all links by radial basis function neural network(RBFNN)and calculates the resource competition ratio of connection requests on the candidate path in both time and frequency domains.Moreover,it periodically updates the weights of all links and recalculates the candidate paths set for the connection requests of different source-destination nodes pairs,after a batch of connection requests is processed in the network.The simulation results show that the proposed algorithm can effectively reduce the traffic blocking probability and improve the network resource utilization.In recent years,due to the appearance of the hotspot cities,traffic between different nodes pairs in the network presents non-uniform.Thus,spectrum converters are placed on some optical switching nodes to reduce the additional increment of the traffic blocking probability caused by this non-uniform traffic.For this network scenario,a RSA algorithm,entitled Spectrum Conversion combined with Prediction(SCP),is proposed.This algorithm still uses RBFNN to predict traffic variations with time on all links.It tries to carry out spectrum conversion at a node placed the spectrum converter,so that the network can accommodate more connection requests when the spectrum resources in the network could not meet the need of a connection request.In placing spectrum converters and allocating spectrum resources,two different combination strategies which are labeled as SCP_R(Randomly)and SCP_BC(Betweenness Centrality)respectively are adopted.The SCP_R strategy randomly selects optical switching nodes to place the spectrum converters in the network,and all connection requests use the First Fit(FF)method to allocate spectrum resource.However,the SCP_BC strategy places the spectrum converters based on the betweenness centrality of nodes.In the spectrum allocation for connection requests,the FF method is tried first to use,and if the allocation fails,the Maximum Path Spectrum Continuity(MPSC)method is adopted after spectrum conversion.The simulation results verify the effectiveness of the two proposed strategies and show that SCP_BC performs better.