[gecode-users] Analysing search performance

[Christian Schulte]

For the time being, the only support you can get is by using a standard C++
profiler. If you are using gcc for compiling just use gprof the see what is
happening. In order to compiler for gprog you have to give -pg as option for
gcc (just check the gcc doc).

Did you at least check the search space? That is, how big is the search tree
with what you have and how big is that in [1]. If you are talking vastly
different trees there is no point in alayzing performance but your model.

Cheers
Christian

--
Christian Schulte, http://www.imit.kth.se/~schulte/ 

-----Original Message-----
From: users-bounces at gecode.org [mailto:users-bounces at gecode.org] On Behalf
Of Lars Otten
Sent: Tuesday, February 28, 2006 12:29 AM
To: users at gecode.org
Subject: [gecode-users] Analysing search performance


Hello everyone,

I've modeled the basic parts of the Tail Assignment Problem [1] in Gecode,
specifically by implementing two custom constraints inverse(x,y) and
tunneling(x,y,z) -- x,y,z being variable arrays of the same length.

I also have access to some of the example instances the author of [1] used
and have tried them on my self-extended Gecode. Unfortunately the running
times for finding a single solution is orders of magnitude worse than what
was accomplished in [1]. The original system is admittedly hand-tailored to
the problem, but then again I've only implemented the basic model, which
lacks some of the more complicated custom constraints the solution in [1]
takes into account.

Hence I'd like to make some kind of analysis of where the depth first search
in Gecode spends it time, to find potential bottlenecks. For a start it
would e.g. be nice to see how much time is spent propagating my custom
constraints (or all constraints in general).

Is there any way to do this in Gecode? I realize this might also be a more
general C++ question, in this case I'd appreciate any pointers on where to
look for more info...

Thanks in advance,
/Lars


[1] http://publications.lib.chalmers.se/records/full_record/6207.html

_______________________________________________
Gecode users mailing list
users at gecode.org https://www.gecode.org/mailman/listinfo/gecode-users