Apples and oranges: a comparison of RDF benchmarks and real RDF datasets