Traffic Analysis And Application Identification In Mobile Network Based On Community Structure

With the development of smart phone and communication tec hnology, the number of applications in mobile network grows fast. In order to deeply understand user behaviors and network manag ement, develop varies of value added services, and improve qualiti es of services and user experience, application identification is one of the most important tasks in network management of mobile ne twork. However, existing application identification methods are mai nly for traditional Internet, but not well studied in mobile network environment. Since mobile network is different from traditional In ternet, this paper studied traffic characteristics of mobile network f rom the perspective of community structure and propose two appli cation identification models.In this paper, based on the real flow data, which we collecte d in GPRS backbone, this paper constructs mobile flow graph to r epresent the interactions between servers and customers. From the perspective of community structure in complex network, firstly, thi s paper thinks each type of application as a community, and then analyzes the basic characteristics of communities, dominated server s, overlap of communities, and Purity of nodes. At last, this paper finds the different and common characteristics among communitie s. For example,1%of server nodes cover more than88%of flow data in mobile network, and Purities of server nodes in communi ty reach at98%.Based on the different and common characteristics among co mmunities, this paper proposes Multidimensional Clustering Identifi cation model. This model can identify the application type of the community, if the network can be divided into communities. Besid es, based on the Server purity, this paper proposes PSI Model (ba sed on Purity and server nodes identification model). PSI model c an fast identify application type of each new coming flow record, and the accuracy can reach at98%.