Power-law revisited: large scale measurement study of P2P content popularity

The popularity of contents on the Internet is often said to follow a Zipf-like distribution. Different measurement studies showed, however, significantly different distributions depending on the measurement methodology they followed. We performed a large-scale measurement of the most popular peerto-peer (P2P) content distribution system, BitTorrent, over eleven months. We collected data on a daily to weekly basis from 500 to 800 trackers, with information about 40 to 60 million peers that participated in the distribution of over 10 million torrents. Based on these measurements we show how fundamental characteristics of the observed distribution of content popularity change depending on the measurement methodology and the length of the observation interval. We show that while short-term or small-scale measurements can conclude that the popularity of contents exhibits a power-law tail, the tail is likely exponentially decreasing, especially over long time intervals.