RE: SPARQL performance for ORDER BY on large datasets

> I'm wondering for some time already if there is any triple store that
> allows to define custom indexes on special predicates or subsets of
> the whole set of triples/quads? All the existing stores I know index
> over all triples in different combinations (spo, pso, ...). Is there
> any research going on towards partial indexes over user-defined
> subsets of triples? E.g. an index over all xsd:dateTime literals.

The Parliament RDF knowledgebase[1] allows you to setup arbitrary indexes. We
use it on our RDF work to do efficient spatial and temporal processing. It's not
as simple as naming the predicate or datatype you want indexed, although it
wouldn't be tough to implement a configurable index that allowed that. We
published a paper[2] for ISWC a couple years ago on how our indexing works.

The open source Parliament download includes the spatial and temporal indexes we
created. The temporal index is based on the OWL Time ontology and will index the
temporal data and appropriately answer Allen Time Interval temporal
relationships, such as before and after. The spatial index uses PostgreSQL and
PostGIS to store spatial data and answer RCC8 spatial relationships.

-- tBs

Troy Self
BBN Technologies

[1] http://parliament.semwebcentral.org
[2] http://www.springerlink.com/index/t00q16g8k9310742.pdf

Received on Thursday, 27 August 2009 11:29:29 UTC