Modeling BitTorrent-like systems with many classes of users

BitTorrent is one of the most successful peer-to-peer systems. Researchers have studied a number of aspects of the system, including its scalability, performance, efficiency and fairness. However, the complexity of the system has forced most prior analytical work to make a number of simplifying assumptions, for example, user homogeneity, or even ignore some central aspects of the protocol altogether, for example, the rate-based Tit-for-Tat (TFT) unchoking scheme, in order to keep the analysis tractable. Motivated by this, in this article we propose two analytical models that accurately predict the performance of the system while considering the central details of the BitTorrent protocol. Our first model is a steady-state one, in the sense that it is valid during periods of time where the number of users remains fixed. Freed by the complications of user time-dynamics, we account for many of the central details of the BitTorrent protocol and accurately predict a number of performance metrics. Our second model combines prior work on fluid models with our first model to capture the transient behavior as new users join or old users leave, while modelling many major aspects of BitTorrent. To the best of our knowledge, this is the first model that attempts to capture the transient behavior of many classes of heterogeneous users. Finally, we use our analytical methodology to introduce and study the performance of a flexible token-based scheme for BitTorrent, show how this scheme can be used to block freeriders and tradeoff between higher-bandwidth and lower-bandwidth users performance, and evaluate the scheme's parameters that achieve a target operational point