Seven Years and One Day: Sketching the Evolution of Internet Traffic
Résumé
It is often stated that 'the Internet is a living beast', constantly evolving with time, the implicit corollary being that its robust and sustainable analysis and modeling are impossible and that obtained results may prove to be outdated before being published. This paper aims at investigating this statement on a scientific basis. A longitudinal evolution study is performed over the traffic collected every day for seven years on a trans-Pacific backbone link (the MAWI dataset). Analyzing this unique dataset enables us to investigate long term characteristics of traffic evolution, both at the TCP/IP layers (packet and flow attributes) and application usage. This provides new insights into central issues, specifically: 'Does link bandwidth increase and statistical multiplexing induce an evolution of traffic towards Poisson, Gaussianity and weaker Long Range Dependence?' Traffic in the MAWI dataset is subject to bandwidth changes, congestions, and to a number of short and long lived anomalies. This allows a comparison of their impacts on traffic statistical properties and yields an overview of traffic anomaly evolution. Also, we show and explain how and why random projection (sketch) based statistical procedures offer an efficient and robust tool to disentangle the impacts of actual long term evolutions from those of time localized events (anomalies and/or link congestions). A study of a 24-hour trace collected under the 'A Day In the Life of the Internet' project on March 19th, 2008 complements these results with an understanding of the typical intraday traffic variability.
Origine : Fichiers produits par l'(les) auteur(s)