We serve all requests through a load-balanced redundant stack of software and hardware components. Thus, should a fault occur in one of the hardware or software components, service can be maintained. We use a virtualised infrastructure.
The SPARQL endpoint comprises two separate software components - the LodeStar web application and the triple store.
- A SPARQL endpoint proxy for the backend triple store. This abstraction layer provides some control over the executed SPARQL queries, along with a graphical frontend.
- A graphical user interface on top of generic RDF data, providing a human-readable HTML-based representation of an RDF resource. The data presented for a resource are essentially the product of a SPARQL DESCRIBE query.
The triple store is currently using an in-memory Virtuoso implementation, backed by a hybrid flash-disk storage array.
The URL resolver is very simple. It handles HTTP requests for data on the rdf.ebi.ac.uk domain, inspects the request headers to determine the content type, and responds with a redirect to the appropriate resource (HTML, RDF-XML or N3). That resource is ultimately served by the LodeStar web application.