On the dependence of internet flow traffic

Internet traffc characteristics have been studied extensively in the last ten years and, nowadays, the Internet community is more conscious about many signifcant traffc properties which are considered vital to statistical models, enabling accurate simulation and prediction of Internet performance. We focus on the structure of the dependence between flow size, duration and mean rate, and show that previous works have proved to be contradictory and misguided because of erroneous statistical approaches. We propose a well established statistical framework based on the theory of copulas, that should be used when dealing with random variables with a general distribution, in particular if heavy-tailed. Our preliminary study confirms known results and, additionally, corrects mistaken methods. Results are based on the proposed statistical analysis on three different network measurement, two from the Internet backbone and one from a GPRS/UMTS access network.