Is there any way to optimize SPARQL queries?

Question

I have unmanaged triples stored as part of individual documents that I am storing in my content db. Essentially each document represent a person, and the defined triple specifies the document URI for the person's manager. I am trying to use SPARQL to determine the length of paths between a manager and all of the people below them in the hierarchy.

The triples in the document look like

<sem:triple xmlns:sem="http://marklogic.com/semantics">
    <sem:subject>http://rdf.abbvienet.com/infrastructure/person/10740024</sem:subject>
    <sem:predicate>http://schemas.abbvienet.com/ontologies/infrastructure.owl#manager</sem:predicate>
    <sem:object>http://rdf.abbvienet.com/infrastructure/person/10206242</sem:object>
</sem:triple>

I have found the following sparql query, which can be used to return a manager, aperson below them in the hierarchy, and the number of nodes distant they are.

select  ?manager ?leaf (count(?mid) as ?distance) { 
  BIND(<http://rdf.abbvienet.com/infrastructure/person/10025613> as ?manager)
  ?leaf <http://schemas.abbvienet.com/ontologies/infrastructure.owl#manager>* ?mid .
  ?mid <http://schemas.abbvienet.com/ontologies/infrastructure.owl#manager>+ ?manager .
}
group by ?manager ?leaf 
order by ?manager ?leaf

This works, but is very slow, even in the case where the hierarchy tree I am looking at is one or two levels deep, around 15s. I have 63,139 manager triples of this type in the db.

Franconian

Shouldn't that be ORDER BY ?leaf as you have only one binding for ?manager. – Pyonephritis

Answer 1

I think the biggest problem is going to be the BIND() - MarkLogic 8 doesn't optimize the pattern you're using at all well. Can you try substituting your constant into the places you use the ?manager variable to see if that makes a big difference? i.e.:

select  ?leaf (count(?mid) as ?distance) { 
  ?leaf <http://schemas.abbvienet.com/ontologies/infrastructure.owl#manager>* ?mid .
  ?mid <http://schemas.abbvienet.com/ontologies/infrastructure.owl#manager>+
    <http://rdf.abbvienet.com/infrastructure/person/10025613> .
}
group by ?leaf 
order by ?leaf

StackOverflow isn't a great place to answer performance questions like this, as it really needs a conversation where we work together to help you. Maybe you can try contacting support or the MarkLogic developer mailing list for this kind of question?

Baalbeer

Without the bind in place, it has a very fast execution. Thanks. – Franconian

Just another comment. It is also fast if I set bind parameters in the sem:sparql call. It is only slow when I bind in SPARQL directly. – Franconian