Increasing data distribution in BitTorrent networks by using network coding techniques

Abstract:
Peer-to-peer networks are well known for their benefits when used for sharing data among multiple users. One of the most common protocols for shared data distribution is BitTorrent. Despite its popularity, it has some inefficiencies that affect the speed of the content distribution. In this paper we investigate methods to improve these inefficiencies by employing network coding. A BitTorrent protocol extension built on top of network coding is proposed that keeps backward compatibility with the basic BitTorrent protocol. With the help of a selfdeveloped simulation environment, we run several measurements of different scenarios. We focus on the initial phase in the distribution of the data, when only one seeder exists along with several leechers. We define a metric for the state of a network to make measurements easily and clearly comparable. Through our results, we show that BitTorrent doesn't behave efficiently at every stage of the content distribution process. By using our protocol extension, this inefficiency can be lowered or completely removed. This decreases the time taken for all peers to collect the data by up to 20%.