A peer-to-peer architecture for distributed and reliable RDF storage

Knowledge management systems share information from multiple sources over the network and may have problems in maintaining the consistency due to node failures and data fragmentation in different locations. In this paper we present an architecture for a load balanced and reliable RDF storage system for semantic information distributed over a peer-to-peer network. Peers are self organized in a ring topology, based on a distributed hash table (DHT), where each node is assigned a segment of the key space that can dynamically change in order to maintain a uniform distribution of the data among the participating peers. Data redundancy is then used to replicate each RDF triple in multiple locations so that, in case of peer failures, neighbour nodes can act on their behalf and return consistent results. Additionally, each node provides an entry point able to resolve atomic, disjunctive and conjunctive SPARQL queries on the network semantic knowledge. The performance of this approach is evaluated by monitoring the effectiveness of the load balancing and redundancy algorithm and the overhead introduced on the network load in both a static (only join events) and dynamic scenario.