A bottom-up model for heterogeneous BitTorrent systems

BT system modeling has great importance for understanding the system performance and improving the protocol design. The challenge of accurate BT system modeling lies in the complicated peer behavior in the dynamic and heterogeneous system. In this paper, we propose a bottom-up model to simplify the performance analysis and increase the modeling accuracy, in which local transitions of the system lead to global macroscopic descriptions by integration. To be specific, we evaluate the download rate of each individual peer by estimating the probability of a connection being made, and the probability of a peer being unchoked, and then later integrate the analysis of each individual peer into the description of the whole system. Such a model characterizes both core parts of the protocol and main features of the system and requires far fewer restrictive assumptions than that used before. By using the model, we provide some interesting insights into setting of parameters, power of free-riding, effect of bandwidth distribution and effectiveness of unchoking. We also validate our model through both simulations and experiments. The preliminary experimental results show that our model is significantly more accurate than existing ones.