PlanetP: using gossiping and random replication to support reliable peer-to-peer content search and retrieval

We introduce the PlanetP system, which explores the construction of a reliable peer-to-peer (P2P) content search and retrieval service using randomly circulated global state between peers of an unstructured community. Our work represents a novel alternative approach to recent P2P systems that focus on enabling very largescale name-based object location using sophisticated distributed data structures. We show that our simpler approach scales to several thousand peers (ultimately targeting the regime of about ten thousand) and converges in several minutes using only modest bandwidth while still maintaining reliable search, ranking, and retrieval similar to an Internet search engine. Unlike current search engines or other P2P systems, however, PlanetP does not require centralized directories or management, nor builds a complex distributed data structure. PlanetP achieves its goals using three major components. First, peers collaborate to maintain local copies of the global membership directory along with compact summaries of shared content using a randomized gossiping algorithm. Second, peers implements a per query, text-based ranking algorithm to help users ignore irrelevant documents. Finally, peers collaborate to replicate unpopular content—popular content is naturally highly replicated via hoarding—using ReedSolomon erasure coding to increase the probability of suc-