A Markov model for the evaluation of cache insertion on peer-to-peer performance

Peer-to-peer file sharing applications generate huge volumes of the Internet traffic, thus leading to increased congestion and costs for the ISPs, particularly due to inter-domain traffic. Thus, analysis of peer-to-peer applications and related optimization approaches (such as locality awareness or caching techniques) has been the subject of extensive recent research. In this paper we introduce and analyze a probabilistic model that employs a Markov chain, aiming to approximate the transient evolution of a swarm with a fixed number of peers. This model estimates the distribution of the number of chunks already downloaded by a certain peer at any time. We also show how this model can serve as a tool to analyze certain properties of peer-to-peer applications, such as monotonicity of performance, and primarily to evaluate the effectiveness of cache insertion in a network serving peer-to-peer. For tractability reasons, the model employs certain simplifications of the original BitTorrent protocol, the impact of which is limited as validated experimentally.