Multiscale analysis and modeling of user session traffic in social networks

Online social networking sites like YouTube are among the most popular sites on the Internet. Understanding these user sessions is important, both to improve current systems and to design new applications of online social networks. This paper reports on the results of a multiscale analysis of user session in traffic traces captured at the edge in campus network. The analyzed traffic is characterized by a non self-similar process for small time scales and strong self-similarity for larger time scales. For smaller time scales, the packet arrival times are independent explaining the weak correlation of the data. For larger timescales the arrival of user session connections can be considered independent. Additionally, we propose a model which allows generating traffic similar to the measured traffic. The model contains only multiscale index allowing influencing all observed changes in LD and variance-time plots. Comparing the output of the model with the real traffic traces we find that the model matches reality accurately.