Distribution of Locations of Peers in BT Swarms

In the past few years, there has been a plethora of research on locality-based approaches to optimize the performance of BitTorrent (BT)-like file sharing system. However, most of these approaches have paid very limited effort to the distribution of locations of peers, which may potentially degrade the performance of BT. The locations of peers consist two identities: Geolocation identity and administrative identity. By measuring and analyzing more than 3 000 .torrent file swarms of BT in 5-month, we find: (1) distribution of locations of peers plays an important role in improving the performance of BT; (2) the types and popularity of languages of the shared files are two important factors affecting the distribution of locations of peers in the file-sharing swarms; (3) peers in the file-sharing swarms are non-uniformly distributed both in country level and AS level, and two propositions about the characteristics of distribution of peers are derived. These two propositions overthrow the assumption based on which many P2P locality-based algorithms are used to optimize traffic, i.e., there are always plenty of peers nearby. Our work shows that our findings can be used to understand and answer questions related to traffic optimization of BT.