To evaluate accurately the performance of the network switching equipment and the scheduling algorithm

it was very important to use a simulated network traffic which represents the major characteristics of real network. The study and analysis on the network traffic in recent years showed the traffic distribution was long-rang-dependent and self-similar

not Poisson process. Combining the modern statistical result on network traffic with the existing traffic models

a practical traffic model was constructed. Based on Pareto distribution and truncated power-tail distribution the model presents nice self-similar property in macro-scale. A practical network traffic generating system is built for evaluating the performance of switch fabric and its scheduling algorithm.