SPARQL for Business
Intelligence

Orri Erling - Program Manager, Virtuoso

For The Data Web to Deliver

• Promise of limitless integration and analysis
• Existing SQL workloads should naturally fall within SPARQL's scope
• Surfing and joining between relational data and RDF should be seamless

What Is Missing?

• Aggregation
• Nested Queries
• Expressions

Mapped vs. Physical RDF

Physical is better, if:

• Predicates are unspecified
• Combining lots off different data sources
• Lots of A-box inference (SameAs, Transitivity)

Mapped is better, if:

• Can push all to RDBMS - about 10x faster
• Data is time-sensitive, frequently changing, very large

When defining mappings...

• Be careful when many tables make one entity - you may get lots of unions
• Make explicit IRI scheme to limit pointless joining
• Know what the mapper and SQL can and cannot optimize

Some Present Work

• Refining generated SQL
• There is no reason why SPARQL mapped should not equal SQL in performance, when going to single DBMS

Joining Between Mapped and Physical

• Special SQL logic is needed when joining IRI IDs of physical quads with IRI strings of virtual triples
• SameAs requires special attention

Use Cases

• OpenLink MIS
  • All accounts, CRM, products, emails, support cases have URIs
• MusicBrainz
• OpenLink Data Spaces
• PHPbb, Mediawiki, Drupal, etc.

OpenLink Software

Thank You!

http://virtuoso.openlinksw.com