Research On The Characteristics Of Mobile Internet Based On Network Traffic Monitoring

With the rapid development of mobile communication technologies, the mobile Internet user scale is increasing year after year. In this scenario, numerous mobile services have become part of people’s lives. For the complex scenes of the mobile Internet, many of the characteristics are not fully understood. With the help of network traffic monitoring techniques, and also Cloud Computing and Data Mining, we analyze the characteristics of mobile Internet, build models on network traffic traces. The major contributions of the paper are as follows.(1)We introduced the Cloud Computing technique into the area of network traffic monitoring. We used Cloud Computing platform to perform data analysis and data storage for the first time. Our Hadoop based network traffic analysis platform supports distributed storage. Users can do distributed computing by uploading scripts. The experiments in this paper also prove that our Hadoop based network traffic analysis platform is very efficient.Data import module, script analyzing module, task implementation module are three major modules in our analysis platform. Data import module is the basis of the platform, which takes charge of uploading network traffic. Script analyzing module interprets scripts language into queries or commands. Task implementation module submits tasks onto the Hadoop framework. The Cloud Computing cluster used in the paper was also described. Based on the network traffic captured at major nodes of China, we proved the high efficiency of the platform.(2) We compared the network traffic characteristics of CDMA network and ADSL network based on the traffic captured at a major node in Southern China. We built an ARIMA traffic model on the two traces, in order to choose proper network traffic samples. The detailed analyses include network protocol distribution, network traffic flow duration distribution, network traffic flow length distribution etc., applications traffic distribution and characteristics of P2P streaming traffic.We found that the average packet length of CDMA trace is smaller than the ADSL traffic trace, indicating that ADSL network can support a better access service. Besides, the download/upload ratio in CDMA network is bigger. The average TCP flow duration in ADSL network is bigger than that in CDMA network, while the standard deviation is smaller in ADSL. Based on the network traffic, we found that the new media applications are taking over the majority of the ADSL network. While in the CDMA network, Web browsing is still in the dominant position. Finally, we conducted research on the peer distribution of P2P streaming traffc, and find that the ADSL and CDMA network flows follow different distributions.(3) We study the MMS (Multimedia Messaging Service) traffic distributions and flow characteristics based on the Cloud Computing platform. It includes traffic characteristics of user scale and different time scale etc. Through the long and short time scale traffic analyses, we analyze the MMS number distribution, transmission time, transmission rate, and build model on the personal MMS inter-arrival time.We use a network traffic monitoring system deployed at the backbone network in Southen China to capture the MMS traffic for a year. In the paper, firstly, we describe the MMS framework and the data capturing procedure. The analyses are divided into long scale and short scale. The former one includes the numbers of MMS transmission per day, MMS protocol distribution, port distribution, Success and Failure rate and content length distribution. The latter one is based on one week MMS traffic, include the personal and non-personal MMS traffic characteristics,2G/3G traffic characteristics, personal transmission rate distributioa At last we build a model on the personal MMS inter-arrival time.(4) In terms of network operating QoS analysis, we propose a cascading network QoS analyzing algorithm based on K-means and C4.5algorithm. The algorithm was testified to be suitable for multiple analyze requirements. Also based on the monitoring data captured from the real Internet, the algorithm was proved to be effective and efficient.The algorithm includes training module and discriminant analysis module. The training module build module based on the historical data, the discriminant analysis module make final decision based on the module. The algorithm processes the data captured from the project "User-oriented Active Network Measurement System". The system is deployed at multiple wireless network accessing points, conducting network QoS monitoring24hours. The paper covers experiments to find a proper K and C4.5discrete value, and use the KKZ algorithm to initialize the cluster center values. Based on the six traces of monitoring data, we compared the performances of cascading network QoS analyzing algorithm, K-mean algorithm and C4.5algorithm. As a result, the cascading algorithm was highly efficient and reduces the noise of single algorithm; also it proved to be suitable to several types of monitoring data.