Efficient peer-to-peer lookup based on a distributed trie

Two main approaches have been taken for distributed key-value lookup operations in peer-to-peer systems: broadcast searches [1, 2] and location-deterministic algorithms [5, 6, 7, 9]. We describe a third alternative based on a distributed trie. This algorithm functions well in a very dynamic, hostile environment, offering security benefits over prior proposals. Our approach takes advantage of working-set temporal locality and global key/value distribution skews due to content popularity. Peers gradually learn system state during lookups, receiving the sought values and/or internal information used by the trie. The distributed trie converges to an accurate network map over time. We describe several modes of information piggybacking, and conservative and liberal variants of the basic algorithm for adversarial settings. Simulations show efficient lookups and low failure rates.