Scalable reconstruction of RDF-archived relational databases

Abstract

We have investigated approaches for scalable reconstruction of relational databases (RDBs) archived as RDF files. An archived RDB is reconstructed from a data archive file and a schema archive file, both in N-Triples formats. The archives contain RDF triples representing the archived relational data content and the relational schema describing the content, respectively. When an archived RDB is to be reconstructed, the schema archive is first read to automatically create the RDB schema using a schema reconstruction algorithm which identifies RDB elements by queries to the schema archive. The RDB thus created is then populated by reading the data archive. To populate the RDB we have developed two approaches, the naive Insert Attribute Value (IAV) and Triple Bulk Load (TBL). With the IAV approach the data is populated by stored procedures that execute SQL INSERT or UPDATE statements to insert attribute values in the RDB tables. In the more complex TBL approach the database is populated by bulk loading CSV files generated by sorting the data archive triples joined with schema information. Our experiments show that the TBL approach is substantially faster than the IAV approach.