Connectivity properties of Mainline BitTorrent DHT nodes

The birth and evolution of peer-to-peer (P2P) protocols have, for the most part, been about peer discovery. Napster, one of the first P2P protocols, was basically FTP/HTTP plus a way of finding hosts willing to send you the file. Since then, both the transfer and peer discovery mechanisms have improved, but only recently have we seen a real push to completely decentralized peer discovery to increase scalability and resilience. Most such efforts are based on distributed hash tables (DHTs), with Kademlia being a popular choice of DHT implementation. While sound in theory, and performing well in simulators and testbeds, the real-world performance often falls short of expectations. Our hypothesis is that the connectivity artifacts caused by guarded hosts (i.e., hosts behind firewalls and NATs) are the major cause for such poor performance. In this paper, the first steps towards testing this hypothesis are developed. First, we present a taxonomy of connectivity properties which will become the language used to accurately describe connectivity artifacts. Second, based on experiments "in the wild", we analyze the connectivity properties of over 3 million hosts. Finally, we match those properties to guarded host behavior and identify the potential effects on the DHT.