Abstract

Peer-to-peer networks, due to their decentralized construction, are a natural platform for anonymous communication and large-scale p2p networks may be the key to widespread deployment of anonymous communications technologies. In order to be scalable, however, p2p networks must maintain a limited view of the network, thereby creating a restricted topology graph of nodes that can communicate with each other. As all communication must follow paths within the graph, we study the information that can be learned about the origin of a path based on observing intermediate nodes. We use both graph models and simulations in our analysis.

In our work, we contrast structured networks, where the topology of the graph follows a mathematical model, and unstructured ones, where arbitrary connections can be made. Unstructured networks often develop an emergent power-law topology; we have found that such topologies are a detriment for anonymity because they have poor mixing (paths remaining correlated to their starting point after a large number of hops) and because the high-degree nodes can be subject to a targeted attack. We show that effective attacks against such networks can be carried out with only a moderate number of compromised nodes and without a global view of the network topology.

Structured networks, on the other hand, tend to have good mixing properties, and de Bruijn networks can be shown to achieve optimal mixing and therefore make an ideal candidate for anonymous p2p networks. We study the approximations to de Bruijn networks used in several p2p systems and show that they provide good anonymity on average, and acceptable anonymity in the worst case, even when the full topology of the network is known to the attackers.