Distributed Reasoning in a Peer-to-Peer Setting

In a peer-to-peer system, there is no centralized control or hierarchical organization: each peer is equivalent in functionality and cooperates with other peers in order to solve a collective task. Such systems have evolved from simple keyword-based peer-to-peer file sharing systems like Napster and Gnutella to schema-based peer data management systems like Edutella [3] or Piazza [2], which handle semantic data description and support complex queries for data retrieval. In this paper, we are interested in peer-to-peer inference systems in which each peer can answer queries by reasoning from its local (propositional) theory but can also ask queries to some other peers with which it is semantically related by sharing part of its vocabulary. This framework encompasses several applications like peer-to-peer information integration systems or intelligent agents, in which each peer has its own knowledge (about its data or its expertise domain) and some partial knowledge about some other peers. In this setting, when it is solicited to perform a reasoning task and if it cannot solve completely that task locally, a peer must be able to distribute appropriate reasoning subtasks among its acquainted peers. The contribution of this paper is the first consequence finding algorithm in a peer-to-peer setting: it is anytime and computes consequences gradually from the solicited peer to peers that are more and more distant. We have exhibited a sufficient condition on the acquaintance graph of the peer-to-peer inference system for guaranteeing the completeness of this algorithm. Our algorithm splits clauses if they involve vocabularies of several peers. Each piece of a splitted clause is transmitted to the corresponding theory to find its consequences. The consequences that are found for each piece of splitted clause must be recomposed to get the consequences of that clause.