[Pellet-users] Concept consistency

[Bijan Parsia]

On Wed, May 30, 2007 6:21 am, bacher at bash-it.de said:

>  Just because i read this,
> 
> 
>> > I'm checking now concept consistency to an OWL file over DIG interface. I
>> > get always java error OutOfMemory on the Pellet DIG Server.
>> > I extended Pellet DIG Server with -Xmx1000m. Now Pellet DIG server is just
>> > checking the concept consistency and but it doesn't stop checking.
>>
>> This can happen.
> 
> i ran into the same problem once. the i recognized that i had to less
> RAM so much of the Harddisk-swap was used by the reasoner and got
> incredible slow. i fixed it moving the reasoner to a computer with more
> RAM and increasing to 3GB. Maybe it's worth a try for you.

So, there are, roughly, two things that can get you into trouble: amount of backtracking (i.e., how much non-determinism you actually deal with) and size of the model/proof you are constructing. These can seriously interact...if each branch you explore gets large and then has to be thrown away...bleah.

It's not too difficult (esp. with inverses) to get "runaway" models. Then you will get an OOM. Sometimes giving it even just a little bit more memory available will cure the problem.

It is possible to stress the GC (or the VM as you pointed out) but that seems to be a somewhat pathological case in general. Worth trying for sure.

I hope I can find someone to pick up the profiling work which should make dealing with this sort of thing easier.

Cheers,
Bijan.