Local Anonymity: A Metric for Improving User Privacy in Tor

In anonymous communication networking, entropy is a popular metric for measuring the average-case difficulty of linking two communicating parties. This paper proposes an alternate view of anonymity, rooted in the observation that global measures of anonymity do not necessarily provide accurate information about the anonymity of an individual user. Such differences can arise due to the characteristics of user needs and local network conditions. We introduce the notion of local anonymity to account for the anonymity of individual users and different sub-regions of the network. Based on this idea, we propose a relay selection algorithm in Tor, introduce a metric for computing the anonymity of individual relays, and discuss the choice of parameters, possible attacks, and defence strategies.