Large-scale distributions of digital content over the Internet require effective algorithms and systems. The traditional BitTorrent model has been successful in the distribution of files on peer-to-peer (P2P) systems for homogeneous users [1,2]. But in the real world, how should voluminous content be effectively distributed to heterogeneous users on diverse connections such as dial-up, digital subscriber line (DSL)/cable, and local area networks (LANs)?
Bram Cohen created the first BitTorrent client protocol in July 2001. However, the debate has continued for years over the best way to efficiently stream voluminous content over networks. In this paper, Liao et al. present two models for ascertaining the performance of BitTorrent-like systems. The first is a steady-state analytical model for predicting the performance of a system with a fixed number of users in a time period. The second is a fluid model that incorporates the transitory behaviors of several classes of users who vacate and connect to a system, to forecast the performance of BitTorrent-like systems.
The authors succinctly critique the deficiencies of the existing algorithms available for coping with BitTorrent traffic systems. They derive algorithms for minimizing download rates and delays for low-bandwidth, medium-bandwidth, and high-bandwidth users. They perform event-driven BitTorrent simulation experiments to gauge the performances of the algorithms in forecasting file download rates and delays. The results reveal that the proposed steady-state analytical and fluid models are reasonably accurate in predicting file download rates and, consequently, delays for categories of users. Clearly, the fluid model is the first of its kind to consider the time-related aspects of any number of user categories when estimating the download and delay rates in heterogeneous settings. Other than the failure to incorporate the variance in the number of neighbors that clients connect to, the fluid model supports effective file sharing and “eliminates the problem of freeriders [that] exploit optimistic unchoking in the original BitTorrent, since such users cannot accumulate any tokens.”