Efficiency of Datalog-backed OWL Lite

Hi, again

Raphael's previous proposal clearly shows how certain primitives of OWL
can be theoretically implemented on top of standard SQL:99-compatible
commercial databases.


However from a practical perspective it is by no means ensured that
an implementation on databases that follows this representation is the
most efficient one, particularly that it is more efficient than a
specialized inference engine such as FaCT, OntoBroker or others.

Usually a given implementation must use further means, such as
materialization
of rules (e.g. such as done for class extensions in Sesame), to achieve
efficiency and scalability. We may even have to restrict the support
of primitives further for certain applications where scalability
is not achievable otherwise.

In order to achieve acceptable performance on
relational databases today, we opted for an even simpler
core language in KAON. Possibly some of you may find the
attached document describing the current status of KAON
and its underlying design rationales interesting.

Best regards,

The KAON team

PS: The current version of KAON does not support OWL yet,
but will provide a native OWL Lite implementation in the
future and provide an OWL Full implementation by
relying on specialized third-party inference engines.