The Efficient Network Traffic Classification Technologies For Heterogeneous Devices

With the increasingly complex Internet environment,the importance of network traffic classification has become increasingly prominent.And it plays an important role in network resource control,network service optimization,network security and other aspects.For example,network traffic classification can automatically identify the traffic from different applications,so as to carry out effective resource allocation and traffic scheduling;Normal and abnormal traffic can be distinguished by network traffic classification,so as to realize effective network intrusion detection.At present,the commonly used methods of network traffic classification include the method based on port number,the method based on deep packet detection and the method based on machine learning.Among them,the network traffic classification method based on machine learning is more common.The early researches mainly focus on the application of traditional machine learning methods,but these methods are restricted by the traffic feature extraction methods,so far there is no recognized traffic feature extraction method in the industry.With the continuous development of deep learning technology,the current research focuses on the construction of deep learning model,which directly takes traffic packets as input to achieve "end-to-end" classification,and uses traffic representation learning to replace the process of traffic feature extraction.However,many parameters in such methods are often given by experience,and the model structure is relatively complex,which affects the practical application.For this reason,this paper studies the problems faced by deep learning in actual network traffic classification,and designs a more effective traffic classification method for heterogeneous devices in different scenarios.Firstly,in a cloud data center network environment,network capabilities are often deployed to high-performance servers.Although HPC resources can effectively support network traffic classification based on deep learning model,there are often multiple applications competing for computing resources on high performance servers.For this reason,we propose a method to dynamically and adaptive adjust the input parameters of the model,and make full use of computing resources to achieve the best traffic classification effect.Secondly,in a simple office network environment,traffic classification functions often need to be deployed to microprocessor-based lightweight routing devices,which have limited computing power and cannot support complex models.Therefore,we propose a traffic classification model based on multiple channel parallel,so as to realize fast and efficient traffic classification on lightweight devices.The prototype system of traffic classification has been successfully developed on raspberry PI device.To sum up,the main work of this paper includes the following three aspects:(1)in order to realize effective network traffic classification on high-performance servers,this paper proposes an online network traffic classification framework based on computational resource perception,which consists of three parts: offline model training,online traffic recognition and convolutional neural network model.The effectiveness of the method was verified by a number of experiments.(2)in order to realize effective traffic classification on lightweight network equipment,this paper proposes a multiple channel classification idea,which divides the data to be processed according to different information fields,enters the convolutional neural network in a multiple channel way,and finally integrates it to obtain classification results.Experiments show that this method can effectively shorten the time of traffic classification.(3)the prototype of intelligent network traffic classification system was realized based on raspberry PI.The raspberry PI was configured as a wireless router,and it can classify the traffic based on the proposed multiple channel classification model,which verified the feasibility of the proposed method in practical application.The experiments in this paper were conducted on the data sets exposed by the domain.The experimental results show that the proposed network traffic classification method is effective and can be used in practical applications.