Search Mailing List Archives
[protege-owl] Query: Performance of SWRL
martin.oconnor at stanford.edu
Sun Aug 15 09:23:50 PDT 2010
The names of the steps are a little misleading. Transforming SWRL rules to Jess does more than simply generating a Jess representation of the rules and relevant OWL knowledge - it also builds a Rete  network for the rules. The final run step simply involves feeding facts through this network. The network construction task is expensive compared to the execution phase in Jess (and in other systems using the Rete algorithm).
I have not done rigorous scalability analyses yet (though have done quite a bit of optimization work) but am actually planning on doing some work on this task in the next few months.
----- Original Message -----
From: "William Fitzgerald" <wfitzgerald at 4c.ucc.ie>
To: "User support for the Protege-OWL editor" <protege-owl at lists.stanford.edu>
Sent: Sunday, August 15, 2010 2:41:30 AM
Subject: [protege-owl] Query: Performance of SWRL
Dear SWRL Experts.
Does anyone have any performance data regarding SWRL?
I notice that SWRL takes a long time when performing the transformation from OWL to Jess
("OWL+SWRL->Jess"). Running Jess ("Run Jess" button) is almost instantaneous after the previous
For example, consider an ontology with 200 instances that will be compared with one another by
examining each of their property relationships. Each individual has 10 property relations in total.
Note, the 200 instances are the domain instances of the respective properties. There are other
instances (range) that form part of the ontology that are not those of the 200 under investigation.
If I write a SWRL rule that compares each individual of the 200 individuals with each other over a
4-tuple of property relations, the "OWL+SWRL->Jess" button will finish its process in a couple of
minutes or so.
However, if I write a SWRL rule that compares each individual of the 200 individuals over all
properties (10-tuple) then it takes much much longer!
Obviously the search space has increased.
Is the problem (time) solely about transforming the OWL model into a Jess model?
Is it a memory problem regarding Protege and Jess in general?
What is the typical or expected scalability of SWRL in your experience when it comes to the number
of individuals being compared/analysed and the number of property relationships by which the
comparison is being made.
Note, I am running the Protege 3.4 GUI, the latest version. Laptop has a Dual Core 2.4GHZ processor
and 3.4 GB ram.
protege-owl mailing list
protege-owl at lists.stanford.edu
Instructions for unsubscribing: http://protege.stanford.edu/doc/faq.html#01a.03
More information about the protege-owl