Unrevealing the structure of live BitTorrent Swarms: methodology and analysis

Abstract

BitTorrent is one of the most popular application in the current Internet. However, we still have little knowledge about the topology of real BitTorrent swarms and how the traffic is actually exchanged among peers. This paper addresses fundamental questions regarding the topology of live BitTorrent swarms. For this purpose we have collected the evolution of the graph topology of 250 real torrents from its birth during a period of 15 days. Using this dataset we first demonstrate that real BitTorrent swarms are neither random graphs nor small world networks. Furthermore, we will see how some factors such as the torrent popularity affect the swarm topology. Secondly, the paper proposes a novel methodology in order to infer the clustered peers in real BitTorrent swarms, something that was not possible so far. Finally, we dedicate special effort to demonstrate that current BitTorrent swarms are experiencing a marked locality phenomenon at the overlay construction level (or connectivity graph). This locality effect is even more pronounced when we consider the exchange traffic relationships between peers. This suggests that an important portion of the BitTorrent traffic is currently confined within the ISPs. This opens a discussion regarding the relative gain of the locality solution proposed so far.