P2P File Diffusion and Interference
The volume of traffic data transmitted over the Internet has enormously increased recently due to the upcoming of peer-to-peer (P2P) file sharing applications. The most popular applications, such as Gnutella, eDonkey, or BitTorrent, are often abused for illegally sharing copyrighted content over the Internet. In P2P technology, each participant (peer) serves simultaneously as client and server which makes the system more scalable and robust and distinguishes it from conventional client-server architectures. However, this also comes at a slight drawback when considering content distribution. Since now, no longer a single trusted server distributes the file, malicious peers (interference peers) can offer fake or corrupted files and disrupt the file dissemination process.
There are two common approaches for the rightful owners of the files to protect their copyrighted property from being distributed. The first is to deliberately introduce files to the networks that are not indexed correctly. Usually, indexing is performed based on the file name, which in the case of such a poisoned file indicates a movie or song title other than the actual file. It is then mistakenly downloaded by other peers and the propagation of the intended file is slowed down. Another well known method is pollution. Here, deliberate corrupt versions of a file are injected to the network, which make use of the simple error correction methods of the file sharing applications. When the received data is recognized as corrupt, it is discarded and newly requested. This delays the overall downloading process and if the downloading delay exceeds the user's patience, he may become frustrated and abort the download.
We use non-stationary simulation techniques to predict the diffusion characteristics of single files shared in a P2P network. With the model we investigate the rate of diffusion, as well as the effects of pollution. We are particularly interested in the influence of these interference peers as they can greatly change the diffusion behaviour of the file. By modifying the population size of interference peers we can achieve approximately the same effects as performing a vaccination of the susceptible population. Additionally, our model takes the distinction between leechers and seeders into account and we show the influence of selfish peers on the file dissemination process. Especially, the ratio between seeders and interference peers and the willingness of the user to share the file are evaluated. The influence of parameters like sharing probability, request arrival rate, or file size on the diffusion process is also investigated.
Hoßfeld, T., Leibnitz, K., Pries, R., Tutschku, K., Tran-Gia, P., Pawlikowski, K.: Information Diffusion in eDonkey Filesharing Networks. University of Wuerzburg (2004).