A performance comparison of Chord and Kademlia DHTs in high churn scenarios

A distributed hash table (DHT) is an important kind of P2P system that provides decentralized services to look up resources in various applications. In this context, Chord and Kademlia are two relevant DHTs and several pieces of work have appeared in the literature where their performance is evaluated. Unfortunately, available results are neither consistent nor concluding. This situation arises from the use of different churn models (i.e., peer arrivals and departures); all of them neglecting the fact that churn happens since the beginning of the lifetime of DHTs. Furthermore, available performance evaluations do not consider that DHT parameter settings are nonequivalent. To address these concerns, in this paper we present an exhaustive, fair and realistic evaluation framework integrated by: 1) A state-of-the-art churn model executed by peers since their creation. 2) An evaluation methodology that considers the difference in meaning of the parameters belonging to different DHTs. 3) A churn metric that quantifies the rate of change of the P2P population. By means of this evaluation framework, we found that under similar conditions, Kademlia exhibits higher performance than Chord. We conclude that DHTs must have mechanisms to deal with high churn during their whole existence, otherwise, they may not achieve a state where peers are correctly connected. Furthermore, our findings suggest that DHTs should rely on less-dynamic peers to improve their performance.