[gecode-users] Slow down between revision 13005 and 13037

[Christian Schulte]

Hi again,

Thanks again for spotting this. The trunk has reverted to the previous
implementation of Region memory management, so performance should be roughly
the same.

I realized that the old (and now current again) implementation of region was
in fact correct. However, its explanation in MPG was as wrong as could be
(actually the example of how to use several regions is now the example of
how to _not_ use several regions) . As I feel so ashamed, I attach the fixed
part from MPG (Jean-Noël: I think this should clarify matters).

Watch out: the trunk has a number of changes that break backwards
compatibility (yes, Gecode 4 is coming closer), you will have to consult the
changelog.

Best
Christian

--
Christian Schulte, www.ict.kth.se/~cschulte/


> -----Original Message-----
> From: Manuel Baclet [mailto:manu.baclet at gmail.com]
> Sent: Monday, September 03, 2012 5:14 PM
> To: cschulte at kth.se
> Cc: users at gecode.org
> Subject: Re: [gecode-users] Slow down between revision 13005 and 13037
> 
> The program is a personal production to find a timetable for students. I
run it on
> a Core i5 laptop with two cores. I have found that a high number of
threads
> increases the chances of finding a solution (from 20% with one thread to
almost
> 100% with 80 threads). I can try to find a simple example if you want.
> 
> Cheers,
> Manuel
> 
> 
> Le 3 sept. 2012 à 16:52, "Christian Schulte" <cschulte at kth.se> a écrit :
> 
> > Thanks, but this is the trunk so this is to be expected in between
releases.
> >
> > Which kind of program did you try? And how many CPUs does your machine
> > have (I guess less than 80)?
> >
> > Cheers
> > Christian
> >
> > --
> > Christian Schulte, www.ict.kth.se/~cschulte/
> >
> >
> > -----Original Message-----
> > From: users-bounces at gecode.org [mailto:users-bounces at gecode.org] On
> > Behalf Of Manuel Baclet
> > Sent: Monday, September 03, 2012 4:42 PM
> > To: users at gecode.org
> > Subject: [gecode-users] Slow down between revision 13005 and 13037
> >
> > Hello,
> > i don't know if you already know about this but i have found a huge
> > slow down between revision 13005 and revision 13037.
> > Testing on the same program with 80 threads on osx 10.8.1 (compiled
> > with clang), the mean time for finding a solution:
> >
> > revision 13005:
> > time 15s
> > cpu system 10%/400%
> >
> > revision: 13037
> > time 50s
> > cpu system 30%/400%
> >
> > Hope this can help, regards,
> > Manuel Baclet
> >
> >
> > _______________________________________________
> > Gecode users mailing list
> > users at gecode.org
> > https://www.gecode.org/mailman/listinfo/gecode-users
> >

-------------- next part --------------
A non-text attachment was scrubbed...
Name: Region.pdf
Type: application/pdf
Size: 49403 bytes
Desc: not available
URL: <http://www.gecode.org/pipermail/users/attachments/20120907/bdf5a7b0/attachment-0001.pdf>