RDFPeers: A Scalable Distributed {RDF} Repository Based on a Structured Peer-to-peer Network

Cai and Frank

rdf semanticweb p2p query search triple store

@inproceedings{cai:www-2004,
  title={RDFPeers: A Scalable Distributed {RDF} Repository Based
         on a Structured Peer-to-peer Network},
  author={Cai, M. and Frank, M.},
  booktitle={13th International Conference on the World Wide Web},
  pages={650--657},
  year={2004}
}

Attempts to provide a distributed RDF triple store in a scalable fashion

Takes each triple and stores it at three hosts

• Each determined by hashing subject, object, predicate respectively
  • URIs and string literals are SHA1 hashed to determine key
  • Locality preserving hash is applied to literal numbers

Uses MAAN---Multi-Attribute Addressable Network---to spread triples

• Extension to Chord to allow multiple keys
• Total messages is O(M log n) to store triple
  • N nodes, M attributes (always 3 in RDFPEER)

Querying is done in obvious way given spreading to three nodes (hashes of sub, obj, pred respectively)

Edutella and successors are very related work

• But based around flooding and super peers, with some routing based on fixed, known schemas

Triples are replicated on neighbors in case of node failure

Popularity is a significant problem in this scheme

• E.g., rdf:type predicates will be plentiful, to say the least
• RDFPeers manages this by tracking how frequent a predicate value is and simply ceasing to index based on it once a threshold is crossed