Design And Implementation Of Network Traffic Identification Method Based On Deep Learning

Network traffic identification can realize traffic classification according to network features,effectively manage traffic and guarantee network service quality.Traffic identification can mark the application type that traffic belongs to,which provides the basis of task scheduling for applications in the cluster.With the expansion of network scale and the rapid growth of traffic,the complexity of network features increases the difficulty of information extraction in the current intelligent traffic identification methods,and the information reduction leads to the decline of identification accuracy.In network traffic identification,deep learning is applied to realize the automatic combination and transformation of complex network features,explore the best representation method,reduce the difficulty of information extraction,and provide the research ideas and solution for improving the accuracy of traffic identification.This paper summarizes the research and typical methods of network traffic identification based on deep learning,and proposes a traffic identification method Deep2FM to improve the accuracy.The representation ability of image is introduced to simplify the complex representation of network features,the traffic is divided into two categories:discrete and continuous features,respectively,using entity embedded encoder and multiple correlation encoder to traverse,so as to reduce the burden of coding on the model.The deep neural network and factorization machine are introduced into the traffic image classifier to realize the parallel interaction of the pixels at the width and depth levels,so as to improve the identification performance.In this paper,real traffic data sets are used for simulation verification,and the results show that this method can improve the accuracy of each category and alleviate the misjudgment in small samples.Furthermore,a network application scheduling method is proposed to solve the scheduling problem of applications in clusters,this method improves the scheduling timeliness to avoid the delay of application response cycle on the basis of realizing the application division by traffic identification.This method abstracts the application scheduling into task scheduling,receives input features that task attributes,the dependencies among tasks and machine parameters to obtain scheduling environment information,builds a deep reinforcement learning model to perceive the dynamic environment,adjust scheduling decisions and optimize scheduling methods.Experimental results show that the proposed method can reduce the optimal time of a task by 35%and reduce the cycle of network application.