Geographical distribution of peers of BitTorrent in multi-granularity scale

Since BitTorrent (BT)-like P2P applications ignore the underlying Internet topology and Internet Service Providers (ISPs) link costs which may lead to huge unnecessary traffic on the Internet and inefficient network resources usages, many locality-based approaches have been proposed to mitigate such stress on ISPs. However, most of these approaches have paid very limited effort to the geographical distributions of peers in BT. Given the scenario that there are very few peers in a client's neighborhood, locality-based approaches will loose the foundation to improve the performance of BT clients. In this paper, we provide a comprehensive picture of the geographical distribution of peers of BT swarms in multi-granularity scale such as in country and autonomous system (AS) level respectively. We investigate more than 2 million torrents with 6 different languages in 12-month period and find: (1) the geographical distributions of peers are found to follow Zipf-like law both in country level and AS level; (2) peers are more apt to clustered into the ASes with higher ranking. These findings provide some new insights for better understanding BT systems.