Dynamic Service Function Chain Deployment Mechanism Based On Node Load Prediction

To solve the increasingly rigid problems of traditional networks and improve the scalability and flexibility of the network,network function virtualization(NFV)came into being.NFV supports the rapid creation,revocation,or migration of network functions(NF),relying on virtual technology to greatly reduce network costs and improve network resilience.Under the NFV architecture,to provide users with precise services and improve the utilization of network resources,and efficient service function chain(SFC)is constructed so that business flows are processed in sequence by multiple NFs in a predetermined logical sequence.At present,researchers are mainly concerned with how to arrange SFC and research the optimal deployment algorithm of SFC.However,SFC has the following characteristics: 1)The flow rate in the SFC is not static;2)The adjustment of the SFC deployment often has a hysteresis.Aiming at the above characteristics of SFC,this thesis proposes an SFC reconstruction method based on node resource consumption prediction and SFC behavior analysis.In delay-sensitive network scenarios,SFC deployment is adjusted in advance to reduce SFC end-to-end delay.The specific work is as follows:First of all,this thesis proposes a method for predicting node resource consumption.This method regards the trend of node resource consumption over time as consisting of two parts: a stable long-term trend and a short-term dynamic change.Decompose historical time series.For time series representing long-term trends,this thesis is based on the prediction model of long short-term memory neural network(LSTM)to predict.For time series representing short-term dynamic changes,a deep Q network model based on virtual actions is proposed.Make predictions.The prediction method in this thesis is compared with the LSTM model,the BP-NN model,and the single-task DBN model.The experimental results show that the prediction method in this thesis is effective and improves the accuracy of the prediction.Secondly,to reduce SFC end-to-end delay,improve user experience,and reduce SFC reconstruction cost,this thesis proposes an SFC behavior analysis method based on the Hidden Markov Model(HMM),which predicts behavior based on whether there is a traffic interval in SFC The future time slot status of SFC can reduce the influence of other interference factors in the network and accurately grasp the reconfiguration requirements of SFC.Finally,combining node resource consumption and SFC state prediction,this thesis proposes an SFC reconstruction decision algorithm,which calculates the end-to-end delay of SFC at the next moment and decides whether to reconstruct it.In addition,to improve the speed of SFC reconstruction,the heuristic algorithm is optimized.Experimental results show that the method in this thesis reduces the SFC reconstruction cost while reducing the SFC end-to-end delay.