| Abstract |
I will evaluate the strong and weak points of the current SWI-Prolog RDF
store (2.0) that was developed in 2004. Since then, the world has
changed significantly. In 2004 we thought that storing 25M triples, with
an escape to 300M on really expensive server hardware, was sufficient to
deal with cultural heritage data, which was acknowledged by winning the
2006 ISWC challenge. Except for some expensive servers, most computers
were still single core. Now, there are literally tens of billions of
linked triples that may contain relevant information about CH objects
and your phone has 4 cores. I will argue that memory-based stores are
still very relevant, but they must concentrate on handling volatile RDF
data that results from crawling and reasoning. I will present version
3.0 of the RDF store. This version is still an unusable prototype, but
it is far enough implemented to be sure that the design can work. I will
also present its features and tentative figures on its performance. |