[gecode-users] High memory consumption during engine declaration

Christian Schulte cschulte at kth.se
Wed Nov 11 14:10:17 CET 2009


Hi,

I don't really know what could be the reason without further looking at the
program. The memory consumed looks massive, actually, almost impossible.

Let's do some math: the search tree has a depth of 45, say with
recomputation possibly using adaptive recomputation that might be around 10
spaces stored during search. That is, each space takes almost 100 Meg! A
thousand variables and 30000 propagators (which propagators do you use?)
should normally take much less.

So, looks really fishy.

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 benoit
Sent: Tuesday, November 10, 2009 11:11 AM
To: Christian Schulte
Cc: 'Mikael Zayenz Lagerkvist'; users at gecode.org
Subject: [gecode-users] High memory consumption during engine declaration

Good morning,

Thank you both for your replies.
> Another hint: if your model uses the standard Gecode search engines, you
can
> inquire how much memory is occupied by the nodes of the search tree. If
> there is a huge difference between what is reported by the search engine
and
> what is reported by your operating system, you program just has a memory
> leak.
>   
I indeed use a standard Gecode search engine (Branch-and-Bound). I 
followed you advice and performed the test: both values are coherent.
> Then, how deep is the search tree? How many variables and propagators does
> your problem have?
>   
Here are the figures:
- Variables: 622
- Propagators: 25654
- Deep: 49

> In the end: we could say much more if you provided a little bit more
detail
> about what you observe. The information that your program uses a lot of
> memory is not that helpful in itself.
>   
Mea culpa: the memory consumption reaches 862MB.

My message was inaccurate yesterday evening: the memory peak occurs when 
the engine constructor is called. It's maybe consistent with the figures 
provided above ?

Benoît

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





More information about the gecode-users mailing list