On 12/02/15 21:21, Richard Davenport wrote:
Hi Bill, you have everything correct. I haven't copied the sdmx or other
ontologies because I assumed that the ontology would be retrieved directly
from it's URI (pointing to the purl.org server).
I'm guessing that I should try downloading that ontology, re-up it to my
own Fuseki, adjust the prefix URI in the query to match and see what
happens? If it works then great. But unless I have fundamentally
misunderstood linked/semantic data, doesn't having to host all of the
definitions and ontologies defeat the purpose?
Think of it as a cache (and incidentally, you don't have the use a
different name in the TDB dataset).
At query time, going off machine is bad:
1/ Performance
It's 200ms ping time from me to California and back (I'm in the UK) and
that's without doing any work at the far end.
A fast, short query is sub 1ms once the DB has warmed up (basically, no
disc access - disc is ~10ms (assume head movement because anytging
predicatable is cached). In fact, more time is going in HTTP overhead
and the logging that actually going the query.
2/ Reliability
What to do if the far end is down?
SPARQL does not go off downloading ontologies just to get the names.
All it needs is the ontology name for the PREFIX and GRAPH to access
local data unless you tell it otherwise with SERVICE.
Looking up an ontology to find a meaning is a different operation, more
likely done by a human and done in small amounts.
Andy
