Abstract
Recent measurements of data communication and telecommunication traffics have shown that network traffic exhibits fractal (self-similar) nature. Such scale-invariant feature is in strong contrast to traditional (e.g. Poisson-like) models of network traffic. Since analytical and empirical studies have shown that fractal traffic can have a detrimental impact on network performance including amplified queueing delay and packet loss ratio, understanding the causes and effects of fractal traffic is an important problem.Excluding effects introduced by user-behaviors, throughout accurate experiments, we show that the influences of “Capture Effect” to self-similarity in Ethernet traffic is crucial in the time scale of interest, especially under heavy loading. It is proved that the heavy-tailed distribution of application data sets is not the only reason for fractal traffic at link level. Transferring bulk files with the same size can also induce self-similarity at link level. And the influences of cross-traffic at link level can play a non-negligible role. We also find that different transport protocols can increase or decrease degree of self-similarity dramatically.