Distributed query processing in the presence of blank nodes

This paper demonstrates that the presence of blank nodes in RDF data represents a problem for distributed processing of SPARQL queries. It is shown that the usual decomposition strategies from the literature will leak information - even when information derives from a single source. It is argued that this leakage, and the proper reparational measures, need to be accounted for in a formal semantics. To this end a set semantics for SPARQL is generalized with a parameter representing execution contexts. This makes it possible to keep tabs on the naming of blank nodes across execution contexts, which in turn makes it possible to articulate a decomposition strategy that is provably sound and complete wrt. any selection of RDF sources even when blank nodes are allowed. Alas, this strategy is not computationally tractable. However, there are ways of utilizing knowledge about the sources, if one has it, that will help considerably.