Research On Network Traffic Prediction And SFC Mapping Algorithm In Network Function Virtualization Scenario

With the in-depth application of Internet technologies of 5G and cloud computing,new network services are constantly emerging.However,the services of the traditional telecom network are rigid,making it difficult to meet the diversified user service requirements and operator networking goals.Therefore,operators urgently need a novel network architecture that can enhance network flexibility and reduce network operating expenses.By using network function virtualization(NFV)technology,operators can deploy the virtualized network functions(VNFs)on standard servers to process service requests and replace expensive equipment with fixed network functions,realizing network reconfiguration and flexible resources allocation on-demand,and optimizing the operator networking goals.In the NFV scenario,operators deploy service function chains consisting of a series of multiple VNFs to meet the user service demands.Supported by the key special sub-project of the National Key Research and Development Program of China named "Broadband Communication and Novel Network"-Research on the network virtualization architecture and verification(2020YFB1805601),this thesis mainly studies the traffic prediction model and resource allocation problems in NFV scenario,with emphasis on the construction of high-precision network traffic prediction model and the SFC mapping problem based on the VNF resource change matrix.Chapter one introduces the research background and value,and summarizes the NFV research organizations at home and abroad,and open source NFV projects.Chapter two mainly addresses the research basis of this thesis,including an overview of the network traffic prediction,the NFV reference architecture and the SFC mapping problem,a summary of the advantages and disadvantages,applicable scenarios and the research status of several network traffic prediction models.It also presents a brief description of SFC mapping algorithms and the related research status from different perspectives.The main research issues and contributions of this thesis are as follows:In the dynamic service application of NFV scenarios,network traffic prediction is beneficial for operators to improve resource utilization efficiency.However,Existing prediction models only consider the temporal or spatial correlation between different network flows,ignoring the combination of the two aspects.Chapter three focuses on the construction of the neural network-based traffic prediction model with adaptive spatial-temporal analysis.To improve the prediction accuracy,this thesis designs a formula for measuring the spatial distance of different network flows,and put forward a new mechanism to consider the spatial-temporal correlation between different network flows for the first time,which is based on the pearson correlation coefficient and the spatial distance of network flows.When predicting the traffic in NFV network,the proposed model predicts the traffic matrixes through the long short-term memory neural network,and obtains the preliminary traffic prediction value of each network flow.Then,the spatial-temporal correlation coefficients are calculated according to the proposed mechanism and then are sorted in descending order,thus several network flows with a strong correlation with the target flow can be selected according to the coefficients.Finally,preliminary traffic prediction values of the selected network flows and the target flow are imported into the back-propagation neural network,and the prediction value for the revised target flow is obtained.The simulation results show that the proposed model significantly reduce prediction error compared to benchmark model.For the static SFC mapping problems,the existing researches only assume that the VNFs have an impact on the bandwidth consumption,without considering the impact on the node resource consumption.Chapter four studies the SFC mapping problem with the VNF resource change matrix,focusing on the impact of the current VNF on the node resource usage of all its subsequent VNFs which belong to the same SFC.To address this problem,an integer linear programming model is formulated with the optimization objective of simultaneously minimizing both the node computing resource consumption and the maximum frequency slot index,and the SFC mapping algorithm based on particle swarm optimization is proposed.The VNF deployment nodes in feasible SFC mapping schemes are encoded as particles,and the mapping schemes are evaluated with the fitness function of the optimization objective.Through the updating iterations,all particles will be continuously adjusted according to the optimal information of the local individual particles and the global optimal information of the particle swarm,thereby gradually converging to the optimal mapping scheme.The simulation results show that the proposed algorithm can effectively reduce the usage of network resources compared with the benchmark algorithm.