KadRTT: Routing with network proximity and uniform ID arrangement in Kademlia

摘要

Distributed Hash Table (DHT) has been widely applied to peer-to-peer (P2P) applications for efficient content lookup mechanisms. IPFS, one of the distributed systems for sharing files, websites, and data on a worldwide scale, adopts Kademlia as one of the DHTs, that is included in libp2p on the network layer of IFPS. Though DHT-based content lookup can scale in the number of peers, how to control lookup performance, e.g., lookup latency and lookup hop count, is one of the issues for a very large-scale network. Thus, it is necessary to guarantee both the lookup latency and the hop count even if the network scale becomes larger. In this paper, we propose a Kademlia alternative, called KadRTT, that reduces both the lookup latency and hop count. KadRTT has two functionalities, i.e., (i) RTT-based lookup target selection not to increase the maximum hop count, and (ii) uniform ID arrangement for each k-bucket to shorten the initial ID distance from content ID. Experimental results by the simulation show that KadRTT outperforms other Kademlia-based DHTs in terms of lookup latency and hop count.