A Longitudinal Study of Small-Time Scaling Behavior of Internet Traffic

During the last decade, many new Web applications have emerged and become extremely popular. Together, these new “Web 2” applications have changed how people use the Web and the Internet. In light of these changes, we conduct a longitudinal study of the small-time scaling behavior of Internet traffic using network traffic traces, available from the MAWI repository, that span a period of eight years. The MAWI traces are affected by anomalies; these anomalies make correct identification of scaling behavior difficult. To mitigate influence of anomalies, we apply a sketch-based procedure for robust estimation of the scaling exponent. Our longitudinal study finds tiny to moderate correlations at small-time scales, with scaling parameter in the range [0.5, 0.75], across the traces examined. We also find that recent traces show larger correlations at small-time scales than older traces. Our analysis shows that this increased correlation is due to the increase in the fraction of aggregate traffic volume carried by dense flows.