[Pellet-users] Non-linear reasoning performance

Kevin Chiu Kevin.Chiu at harmonicinc.com
Fri Jun 1 08:03:26 UTC 2007


I am new to this Ontology stuff. I have been running some benchmarking
on reasoning with Pellet in Jena.

What I mainly want to do is classification on some data instances. And
there are several datatype and object properties over the classes and
restrictions on defined classes (e.g like name owl:hasValue "a" -> Class
a).

 

And here are the benchmark results got on a 1GB ram 3GHz P4 desktop,
each is got by an individual program run:

 

->< 34 classes, 4 properties, 90 individuals >

->Reasoner get instances           : 0.78 Seconds

 

->< 34 classes, 4 properties, 140 individuals >

->Reasoner get instances           : 0.391 Seconds

 

->< 34 classes, 4 properties, 235 individuals >

->Reasoner get instances           : 2.609 Seconds

 

->< 34 classes, 4 properties, 440 individuals >

->Reasoner get instances           : 9.859 Seconds

 

->< 34 classes, 4 properties, 665 individuals >

->Reasoner get instances           : 46.891 Seconds

 

->< 34 classes, 4 properties, 2540 individuals >

->Reasoner get instances           : 999.781 Seconds

 

<******takes too long, aborted in the middle******>

->< 34 classes, 4 properties, 9430 individuals >

->Reasoner get instances           : >45300 Seconds

 

My questions are:

1)       The reasoning performance is quite poor, compare to figures
I've seen on some papers. I've tried changing the restriction on object
properties instead of on literal string, it goes even much more slower.
Is there anything reference on the limitations or recommendation on
writing OWL (or using Pellet) to avoid performance hurt?

2)       The time-cost vs instances count is non-linear, is this normal?
I've heard something about incremental reasoning but seems it won't come
out till Pellet 1.5, right? And will it help if I store an inter-media
graph from small data set onto persistent DB then add instances to it?

 

Thanks in advance.

Kevin Chiu.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.owldl.com/pipermail/pellet-users/attachments/20070601/7f4adc7c/attachment.htm 


More information about the Pellet-users mailing list