Research on the models of the distribution of files on the networks

This article describes some models of the distribution of files on the networks, especially the Pareto distribution model and lognormal distribution model. Simulations on true network files demonstrate that lognormal distribution is closer to the true data sometimes, especially in the tail part. The traffic generated by the files with lognormal distribution also shows self-similarity, which suggests self-similarity is not only due to Pareto distribution of file sizes, files with lognormal distribution can also lead to self-similarity of traffic. This conclusion may lead to some changes of file size models in simulation work, which will cause different system design.