Analysis of Content Distribution Architectures in P2P Systems

This paper takes a novel perspective to P2P networking. Can a P2P system be used for the distribution of files with time-critical data such as relevant software patches or virus footprints updates? We examine and compare different distribution architectures based on linear and tree topologies built on top of the P2P overlay, including in the analysis the presence of heterogeneous bandwidths, both symmetric and asymmetric access links. We propose an analytical solution of the distribution process that not only yields the mean download time but also the distribution of the download times. We validate the analytical model against a Monte Carlo based numerical solution, which can also be used to analyze scenarios where correlation and dynamic behavior make the theoretical analysis too approximate. The insights we gain are used to devise modifications of the distribution strategies that achieve good performance even when slow access links and incomplete knowledge jeopardize the fast delivery of the content. Indeed, with proper (yet simple) mechanisms the average completion times achieved are close to the minimal (lowest possible) ones, which indicates that the P2P paradigm can be used for delivering time-critical data. Finally the presence of non-cooperative peers is analyzed, assessing their impact for different scenarios.