Technical Report(Oct 2017)
Keywords: network traffic; DHT
During recent years, Distributed Hash Tables (DHTs) have been extensively studied by the networking community through simulation and analysis, but until recently were not adopted by popular P2P applications. Therefore, it was not feasible to examine DHT behavior in practice. Recently, the popular eMule file-sharing software incorporated a Kademlia DHT, called Kad, into their software. The success of Kad appears to have triggered other P2P applications (notably BitTorrent) to also incorporate Kademlia.
In this paper, we empirically study the performance of the key DHT function, lookup, over this DHT which has more than one million simultaneous users. We develop new analysis for predicting performance, characterize the accuracy of Kad's routing tables, and empirically determine the optimum level of parallel lookup and the level of replication needed to cope with lookup inconsistencies. Our results show that compared to the current eMule client, overhead can be significantly reduced while simultaneously increasing performance.