An Empirical Performance Evaluation and Modelling of BitTorrent peer-to-peer File Sharing System using Queuing Network Models

It has been observed that the maximum bandwidth usage incurred during the file transfers on the internet mainly corresponds to the large chunks of data, which typically consists of continuous media, for example videos, movies or the sound files. Data packages like bulky software executables also contribute to the bandwidth usage. Peer to peer file transfer systems have now become enormously popular on internet due to several advantages. One of the main advantages is the reduced over head on one centralized entity as the file transfer process is distributed. However some of the traditional file transfer systems do not scale well with the number of clients. BitTorrent is an improved peer-to-peer(P2P) file distribution protocol. Several free implementations of this protocol exist. This protocols has various mechanisms to award the peers who are actively participating in the file transfer process. These incentive mechanisms and various other mechanisms make BitTorrent scalable and robust. We would investigate these aspects in detail in the later chapters. In this project we have tried to understand the BitTorrent protocol and have performed an experiment to understand the performance and incentive mechanism imbibed in the protocol. The characteristic graphs are studied to find out the patterns of similarity and relationships between seemingly disjoint parameters. Through the medium of empirical observation we have tried to characterize various parameters of the BitTorrent system so that it could be modelled using a queuing network model. We have also tried to find the asymptotic bounds on the performance. The study is further purported by the comparison between the simulation and experimental results mentioned towards the end.