Selecting Materialized Views for RDF Data

In the design of a relational database, the administrator has to decide, given a fixed or estimated workload, which indexes should be created. This so called index selection problem is an non-trivial optimization problem in relational databases. In this paper we describe a novel approach for index selection on RDF data sets. We propose an algorithm to automatically suggest a set of indexes as materialized views based on a workload of SPARQL queries. The selected set of indexes aims to decrease the cost of the workload. We provide a cost model to estimate the potential impact of candidate indexes on query performance and an algorithm to select an optimal set of indexes. This algorithm may be integrated into an existing SPARQL query engine. We experimentally evaluate our approach on a standard query processor. We claim that our approach is the first comprehensive suggestion for the index selection problem in RDF.