A logical characterisation of SPARQL federation

摘要

The paper provides a logical characterisation of distributed processing of SPARQL queries. The principal notion analysed is that of a distribution scheme; a pair consisting of an evaluation rule and a distribution function. Different choices of evaluation rules and distribution functions give different federation schemes that are proved to be sound and complete wrt. different selections of RDF datasets. Three distribution functions are singled out for special attention, called the even-, standard- and prudent distribution respectively. All yield sound and complete federation schemes when combined with an evaluation rule named the collect-and-combine rule. The completeness results thus obtained are next compared to existing federation systems with the aim of illustrating the potential utility of a framework in which salient properties can be formulated and explored. The final section, relates distribution schemes to the heuristic notion of a join-ordering to yield sound and complete execution plans for the aforementioned federation schemes.