Is CoV(t)-based Modeling Sufficient for Traffic Characterization in Network Links ?

For performance evaluation and dimensioning of packet-based networks, engineers need simple, efficient and realistic traffic models. The traffic volume on a packet link, observed at different time scales t, has previously been modeled as a stationary stochastic process based on the Coefficient of Variation CoV(t). In this paper we try to supply the missing information about the shape of the distribution functions, required to fully characterize the traffic in network links. Applying the maximum entropy principle, we show that the knowledge about the mean rate μ and the value of CoV(t), associated with the information about the link capacity C, gives truncated Gaussian distributions that are bathtub-shaped at short time-scales and bell-shaped at long-time-scale, leading to quite accurate modeling of the traffic volume distributions at all time-scales t. We illustrate this with some real traffic and some simulated traffic. Keywordsnetwork dimensioning; traffic modeling; traffic simulation; coefficient of variation;