[gecode-users] Slow performance on large domains before branching

Guido Tack tack at gecode.org
Sun Sep 20 12:52:37 CEST 2015


Hi,

sorry for the late reply.  This does look a bit odd, apparently the set-union propagator needs to do a lot of work at the root node.  It’s a bit difficult to say what’s going on without looking at the code.  One possible way to improve the performance is to check whether there is some simple preprocessing you could do, to make sure the domains of the set variables are small to begin with.  But there might be something wrong with our set union propagator.

I would also be happy to have a look at your code directly, if it’s not too difficult to share (just send me an email).  Or perhaps you can extract an example that shows the same behaviour (it looks like all we need are the set variables and constraints, your custom brancher isn’t part of the problem).

Cheers,
Guido

-- 
Guido Tack
http://www.csse.monash.edu/~guidot/



> On 31 Aug 2015, at 4:46 pm, Matthias Goldhoorn <matthias.goldhoorn at uni-bremen.de> wrote:
> 
> Hello Developers and Users,
> i have a quite big domain/searchspace.
> However gecode performs currently well and solves my problem. But i recognized a strange behavior.
> 
> Berfore ma actual branching/search starts, gecodes does a lot of internal calculations. I'm not sure if i have done the setup wrong of if there is some optimization that i can do?
> I attached [1] a picture of the most time and instruction consuming calls. Maybe somone can give me a recommendation where i can start tuning..
> 
> Setup (class instanciation and bracer setup ~15 sek)
> first call of BAB until code-flow reaches my first (self written) brancher: ~3min
> Actual branching/searching: 10sek.
> Overall Memory consumption ~5GB
> 
> Best Regard,
> Matthias
> 
> [1] https://www.dropbox.com/s/21aqlbnjs3zx1rd/callgrind-gecode.png?dl=0
> 
> -- 
> Dipl.-Inf. Matthias Goldhoorn
> Space and Underwater Robotic
> 
> Universität Bremen
> FB 3 - Mathematik und Informatik
> AG Robotik
> Robert-Hooke-Straße 5
> 28359 Bremen, Germany
> 
> Tel.:     +49 421 178 45-4193
> Zentrale: +49 421 178 45-6550
> Fax:      +49 421 178 45-4150
> E-Mail:   matthias.goldhoorn at uni-bremen.de
> 
> Weitere Informationen: http://www.informatik.uni-bremen.de/robotik
> 
> 
> _______________________________________________
> Gecode users mailing list
> users at gecode.org
> https://www.gecode.org/mailman/listinfo/gecode-users




More information about the users mailing list