On 04/09/13 11:32, Diogo FC Patrao wrote:
Hi Andy
Thanks for answering! I'll look into OpExecutor to see if I get somewhere.
Couldn't find the webpage to this quack library, is it published yet?
Not yet; its in my github area (user 'afs')/
There is code there that might help you (that said, hash join is
actually one of the simpler join algorithms to implement if you want
in-memory joins datastructures).
Andy
Cheers,
Dfcp
On Tuesday, September 3, 2013 <x-apple-data-detectors://24>, Andy Seaborne
wrote:
On 02/09/13 19:03, Diogo FC Patrao wrote:
Hi
I'm running a query with join results from two endpoints:
SELECT * {
SERVICE <s1> { ?a a :Class1 } # q1
SERVICE <s2> { ?a a :Class2 } # q2
}
I noticed (by running ARQ 2.8.8 with -v) that it first dispatches query q1
on s1 and then one query q2 on s2 *for each result* in the previous query.
Is there any other way of doing it? I mean, I'm getting like 1M results
from q1, maybe it would be better to get all results from q2 and then join
them in memory, or pass more than one value at a time for q2 (the VALUE
tag
allows that).
Without programming, there is not a way to control the execution. You can
implement a OpExecutor and provide your own implementation of join.
There are many issues such as how to know the sizes from a remote service
and how to send data to a remote service from s1 to s2 -- VALUEs assumes
looping back through the query engine, but does s2 allow huge SPARQL. See
bind joins from the IBM Garlic papers.
ARQ provides the basics - remote execution - but isn't a federated query
optimizer.
There is a not-yet-ready query engine, called 'quack', that provides merge
and hash joins - only on BGPs currently but the code, in principle, is
general.
Andy
Cheers!
--
diogo patrão
