Lookup Parameter Optimization for Kademlia DHT Alternative in IPFS

Peer-to-peer networks, such as IPFS, adopt the distributed hash table (DHT) to efficiently find contents. In particular, Kademlia, which is used in IPFS, requires parameters regarding the lookup concurrency, the number of next hops, and the k-bucket size. However, such values are manually set and then the configuration is not optimal for minimizing the network latency in any network dynamics. In this paper, we present a method for automatically deriving the optimal lookup parameters for KadRTT, which is a modified version of Kademlia to improve the lookup latency. We derive the optimal values for the k-bucket size, lookup concurrency, and the number of next hops using the lookup message arrival rate, initial ID distance, and lookup iteration count. From the experimental comparisons by both a simulation and an emulation, we show that our proposal contributes to the lookup latency, and overlay hop count.