Re: [Help-glpk] [Fwd: memory]

[Mike Steglich]

Hi Ronald,

I'm from the CMPL team and I would like to analyze your problem. It could be - as mentioned by Xypron - a memory problem.

Which version of CMPL do you use? Is it a binaries from the CMPL website or did you compile CMPL for your own? If so, did you compile it in 32 or 64 bit mode?

Is it possible to send me your CMPL model? You can be sure that i would use it only for analyzing the problem.

Cheers

Mike 

Am 25.11.2011 um 15:20 schrieb Ronald Wolthoff:

Thanks for your responses. Yes, GLPK is the solver and CMPL the preprocessor. I meant that the increase in preprocessing time seemed steep relative to the increase in solving time, but of course these are different processes, so there is no obvious reason why they should behave in the same way.
Thanks for the hint on sparse problems and direct access to the library. That seems promising and I will look into that.

Nigel Galloway

November 25, 2011 8:32 AM

Do you mean GLPK is Solver and CMPL is preprocessor?

What do you mean by stronger than expected? What happens at 29 and 31.
Draw a graph of times between 15 and 30. I would say that, if the graph
is smooth then the value for 30 is exactly as expected. If its smooth to
29 then discontinuous to 30 it is unexpected.

Andrew Makhorin

November 24, 2011 4:11 PM

-------- Forwarded Message --------
From: Ronald Wolthoff <address@hidden>
To: address@hidden
Subject: memory
Date: Thu, 24 Nov 2011 09:44:57 -0500

Hello,

I'm using GLPK in combination with CMPL and I encounter some issues with
the time required for preprocessing. My problem has 2N^2 rows and 2N^4
columns (real variables, less than 1% non-zeros). When I run the problem
for N=15 (450 rows, 101250 columns), the preprocessing takes
approximately 1 minute and the solver 3 seconds, which is great. When I
use N=30 instead (1800 rows, 1,620,000 columns) , the solving time goes
up to a few minutes, which is still great, but the preprocessing time
explodes to something like 9 hours. Although the problem is fairly
large, this increase seems stronger than expected. Is there a way I
could reduce the preprocessing time, e.g. by allocating more memory to
the program?

Many thanks,

Ronald




_______________________________________________
Help-glpk mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/help-glpk

_______________________________________________
Help-glpk mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/help-glpk