federated querying with lightweight interfaces

Dear all,

In a recently accepted JWS article [1], we evaluated the Triple Pattern Fragments (TPF) interface
in different ways, which includes a federated scenario (FedBench + added complex queries):
http://linkeddatafragments.org/publications/jws2016.pdf

Especially the performance in the federated scenario proved rather interesting:
both for completeness and execution time, the TPF client over the public Internet
achieved scores similar to well-known SPARQL federation engines over a closed network
within the measured interval of 300 seconds. For some queries,
the TPF client achieved better completeness and/or shorter query execution times.
Note that the TPF client did _not_ perform a separate source selection step in advance,
in contrast to other engines, challenging conventional wisdom about query federation.

This is an indicator that lightweight interfaces are more than just an academic exercise.
Especially in federated scenarios, which in my opinion show the true value of Linked Data,
they can provide a viable and cost-effective solution to publish data—
as long as we remain realistic about the kind of queries we expect to yield answers fast.

Furthermore, TPF is just one of the many lightweight interfaces we can imagine:
several others are possible (some of which are subsets of SPARQL).
And if we make interfaces self-describing, as we do with the TPF of interfaces,
clients can automatically discover what features servers have, and adapt accordingly.

Our results confirm that data does not need to be either a SPARQL endpoint or a dump:
even a very limited query interface can deliver promising results in a federated setting.
As such, this mail is in the first place a plea for diversity:
making querying work on a Web scale can benefit from lightweight interfaces.

Best regards,

Ruben

PS For those who would like to experiment with live federated queries,
see: http://bit.ly/swedish-nobel-harvard and http://bit.ly/cubist-works

Received on Wednesday, 4 May 2016 13:18:52 UTC