Research On Carrying Performance Of Multi-service In Large Capacity Access Network

With the rapid growth of data traffic in the home network service of the current access network,the application often seizes resources to achieve the actual carrying effect for the limited network resources.The differentiation of services in the network timely and accurately is of great significance for improving network performance and ensuring service quality.The multi-service carrying of the home has become a key link in the entire service carrying network.The service traffic of wireless access in the home network has the characteristics of large data volume,encryption and uneven distribution of data packet length.The service carrying system needs to quickly classify and control the traffic in real time.The traditional service carrying design mainly improves the bearer performance by limiting the traffic generated by the device according to the MAC address of the device.However,the service carrying system can control the traffic of the specific service type application by identifying the traffic of the mobile terminal application.Finally,it is not limited to controlling the traffic of a certain device and the bandwidth control sinks to the application side.In order to ensure better multi-service carrying performance after the emergence of large data traffic,this thesis analyzes the time characteristics and spatial characteristics of mobile application traffic.The whole experiment platform system is constructed on the Raspberry Pi,which is based on Convolutional Neural Network(CNN)and Long Short-Term Memory network(LSTM)in order to study the carrying performance of mobile terminal application in this system.So the specific research work and contributions include:The spatial characteristics of network data flow are studied.Based on the basic principle of CNN,the service carrying system on the Raspberry Pi platform is designed and constructed.The CNN network classification model and real-time flow control model are established in this system.The training data set suitable for CNN network classification model is designed.The influence of network flow length,CNN network training parameters and model parameters on classification results will be analyzed.After that,the control performance of the real-time traffic on this system is considered.The time characteristics of network data flow are studied and the basic principle about time series processing by LSTM(Long Short-Term Memory)network is analyzed.The LSTM service carrying system is designed based on LSTM network on Raspberry Pi,where the LSTM network and data flow control model are constructed.According to the requirements of the LSTM network for detecting data,a method by converting the original stream of the LSTM network into a training set is designed,so that the system can identify and control the current stream in real time through the temporal characteristics of the stream.At the same time,the influence of the length of network flow sequence,the length of sequence elements,LSTM training parameters and LSTM model parameters on classification results,is analyzed.Moreover,the control performance of the data flow on this system is considered.