Peer Clustering for the InterPlanetary File System

Abstract

Distributed Hash Tables (DHT) are once again attracting research interest, 20 years after their inception, as a promising solution for scalable and decentralized content storage. A prominent example is the InterPlanetary File System (IPFS), a distributed peer-to-peer storage system with more than 20K online peers, which uses the Kademlia DHT. We design and implement H-Kademlia, a hierarchical version of Kademlia for IPFS, where peers are distributed into disjoint sets of users, or clusters. Peer clustering can offer resilience to network partition, privacy of in-cluster content lookups, as well as improved caching, content filtering and access control. We assess the performance of IPFS over H-Kademlia via simulations that use real traces from the IPFS network. Our findings show that our design delivers the benefits of peer clustering, without significant performance penalties.