Sebastian Kaune, TU Darmstadt

At present, the probability of selecting ``the peer next door'' as an overlay neighbour in Distributed Hash Tables is fairly small. Prior research has been concerned with reducing the lookup latency by means of proximity neighbour and route selection, but focused on recursive routing algorithms. This talk focuses on the widely deployed Kademlia DHT, due to its iterative routing. We introduce an underlay metric capable of providing information about the underlay network to the algorithms of the overlay network. The goal of an underlay metric is to optimise routing according to the needs of a specific application. For instance, routing could be optimised to maximise the locality of traffic, reduce the latencies of lookups, prefer communication with peers that have high-bandwidth connections, or avoid routing to untrustworthy peers. To this end, we present underlay metrics which leverage location data about peers and extends Kademlia’s iterative routing algorithm to reduce cross-network traffic at the level of the distributed hash table. Evaluation with real-world measurement data gives evidence that locality of traffic tends to reduce lookup latencies as well. In turn, mechanisms that aim at reducing lookup latencies do not necessarily reduce cross-network traffic to the same extent.