Research On Multi-objective Optimization Model And Intelligent Algorithm Of SDN Controller Deployment

With the rapid development of mobile communication technology and the popularity of Internet related applications,the current information network is developing toward large capacity,high bandwidth,high reliability,low latency and wide coverage.Due to the help of software defined network(SDN)technology,network equipment manufacturers and operators can achieve significant increases in optical network capacity and bandwidth,flexible expansion of network architecture and dynamic demand-based provisioning of network resources in the optical networks with dense wavelength division multiplexing(ON-DWDM).When ON-DWDM is improved through SDN,the number and location of SDN controllers in the optical network must first be determined,which is usually referred to as SDN multi-controller placement problem(SDN-MCPP),and is one of the current research hotspot of SDN.To meet both the diversified operator and user needs in the differentiated network environments,controller placement algorithms must provide a set of controller placement schemes in which objective values do not dominate each other.This thesis studies the modeling of SDN-MCPP and its multi-objective intelligent algorithm in ON-DWDM.The first two chapters of this thesis briefly summarize the traditional optical networks,ON-DWDM with corresponding research organizations at home and abroad.It introduces ON-DWDM components and architecture,SDN architecture and key technologies as well as the advantages of applying SDN to ON-DWDM.The second chapter also discuss SDN-MCPP and its research status,and focuses on the multi-objective optimization model and its solving algorithm in SDN-MCPP.The main work and contributions of this thesis includes the following two parts:(1)The construction of the SDN-MCPP multi-objective model and the design of its multi-objective intelligent solving algorithm.This thesis firstly analyzes and summarizes the existing SDN-MCPP algorithms,most of which are heuristic algorithms,and there are few intelligent algorithms that consider both deployment cost,load difference and propagation delay.In this thesis,a SDN-MCPP three-objective optimization model is established in ON-DWDM,while simultaneously optimizing controller deployment cost,controller load difference and control network propagation delay.To solve such a model,a binary coding scheme coding scheme is designed,and an multi-objective hybrid evolution with information entropy awareness(MOHEIEA)is proposed.The features of this algorithm include: a)uses a hybrid algorithm to generate initial population,which combines clustering algorithm and uniform initialization method;b)the mechanisms of encoding conversion,information entropy awareness and hybrid evolution are specially designed for the coding scheme to generate solution set.Besides,it adaptively adjusts the probabilities of crossover and mutation in the population according to the different evolution stages;c)after the evolution of each generation is completed,the perturbation modification mechanism is used to correct the individuals with higher deployment costs in the population.The proposed algorithm fuses the multiple mechanisms mentioned to both improve the algorithm convergence and increase solution diversity.Finally,the algorithm performance simulation is completed in the network topologies of Internet OS3E?Interroute and Cogentco.The results show that the proposed algorithm can obtain more dominant and diverse solutions than those of the baseline algorithm to a certain extent,and it is suitable for networks with different scales,thus validating the effectiveness and universality of the proposed algorithm.(2)The construction of the SDN-MCPP many-objective model and the design of its many-objective intelligent solving algorithm.In this thesis,a five-objective optimization model is established based on the work in the third chapter,while aiming at minimizing controller deployment cost,controller load difference,propagation delay between controllers,propagation delay between controllers and switching nodes,and unreliability of the control network.To solve such a model,the improved elite selection strategy and solution set distribution maintenance strategy are added to the many-objective hybrid evolution with information entropy awareness-extended(MOHEIEA-E).The features of this algorithm include: a)selects the elite individuals in the population through loose non-dominated sorting;b)estimates the distribution of individuals according to the Harmonic average distance.MOHEIEA-E fuses the multiple mechanisms mentioned to increase the sovling ability of the algorithm optimization.Finally,the algorithm performance simulation is completed in the Cogentco network topology.The results show that MOHEIEA-E can obtain more distribution and convergence solutions than those of the baseline algorithm,thus validating the effectiveness of the improve algorithm.