Re: Efficiency/scalability

Interesting.

I spent a bit of time trying to find out more about this, but was unable
to find anything except the presentation
http://colab.cim3.net/file/work/SOACoP/2007_05_0102/BThompson05022007.pdf
and, of course, the code on SourceForge (with no documentation that I
could find).

I wonder if there is any good documentation.  It might be useful for
someone who cares a lot about this sort of thing to try to dig some
information up.

Peter F. Patel-Schneider
Bell Labs Research


From: Jim Hendler <hendler@cs.rpi.edu>
Subject: Efficiency/scalability
Date: Tue, 18 Dec 2007 23:12:19 -0500

> One thing that has come up in the fragments discussion has been the  
> whole issue of provable properties vs. real-world scaling (sometimes  
> called theoretical efficiency vs effectiveness) - question were  
> raised about the theoretical properties of some of the RDFS 3.0  
> stuff, where I can only say, we're still exploring this, but I note that
> 
> http://www.bigdata.com/projects/multiproject/bigdata-rdf/index.html
> 
> reports on the handling of  "entailments ...for RDF Schema,  
> owl:sameAs, owl:equivalentProperty, and owl:equivalentClass" at  
> speeds that are pretty amazing (load at 21,000 triples per second,  
> compute at 8100 entailments for second in computing the RDFS+ closure  
> for Wordnet) and in mail on the billion triples mailing list they've  
> proposed that we up the challenge to 10B triples to make things  
> challenging...
> 
> i realize they are still far less expressive than the RDFS 3.0  
> proposal (or any of our fragments) and they have no negations and use  
> the realized triples trick to create a finite universe - it's just  
> that 10^9 is a pretty big finite universe, and it's important to  
> realize that RDF DBs are reaching those sizes already - and including  
> some RDFS and OWL constructs, so the Abox stuff is really getting  
> impressive (and hard to ignore)
>   -JH
> p.s. this is no means meant to endorse bigdata-rdf, a project I know  
> nothing about beyond what is on that web site.
> 
> Prof James Hendler				http://www.cs.rpi.edu/~hendler

Received on Wednesday, 19 December 2007 10:16:26 UTC