[Cbc] Performance

[Anthony Presley]

John,

Thanks for getting back to me.  We are content to stick with cbc 1.2,
and as an organization, do not have the "domain knowledge" (with MIP/LP
problems) to be able to diagnose the problem very far.  We can certainly
compile code, run a test, and report numbers :-)

I would be happy to provide you (or another contributor) with some data
sets that we are using, if that would be of use for you.

Thanks for your help!


--
Anthony

On Sun, 2007-12-09 at 05:58 -0500, John J Forrest wrote: 
> Anthony,
> 
> Unlike for LP's,  tiny changes to the code can make a large difference in
> the running time for MIP's.  I have had cases where what seemed to be a
> trivial change has increased the runtime for one problem by a factor of a
> 100 - without any logical reason that I could see.
> 
> However that is not very helpful to you.  Compiling without threads should
> make no difference.  Without more information it is difficult to help.  I
> would try allowing strong branching and maybe playing with
> "trustPseudoCost"
> 
> John Forrest
> 
> 
>                                                                                                                                              
>   From:       Anthony Presley <anthony at resolution.com>                                                                                       
>                                                                                                                                              
>   To:         cbc at list.coin-or.org                                                                                                           
>                                                                                                                                              
>   Date:       12/07/2007 01:15 PM                                                                                                            
>                                                                                                                                              
>   Subject:    [Cbc] Performance                                                                                                              
>                                                                                                                                              
> 
> 
> 
> 
> 
> All:
> 
> I just thought I'd post some performance data we saw with a data set.
> It's an integer minimization problem.  I downloaded and tried out cbc 2
> (in comparison to cbc 1.2), and ran the model with these parameters (in
> both versions):
> 
>              import file
>              sec 82
>              allow 82
>              strong 0
>              threads 3
>              solve
> 
> I also tried with threads -1, 1, 2, 3, and 0.
> 
> Overall, the end result is that cbc 1.2 takes 30 - 32 seconds, and cbc
> 2.0 takes about 78 - 80 seconds.
> 
> The only difference that I know of (besides the version and features in
> the software) is that cbc 2.0 was compiled with
> --enable-cbc-parallel  .... ie, I so no speed improvement with parallel
> enabled.
> 
> I haven't tried cbc 2 without the "--enable-cbc-parallel".
> 
> We will continue using CBC 1.2, but just wanted to see if there was
> anything we might want to look at that may be slowing down the solution.
> Did some of the defaults change?
> 
> 
> --
> Anthony
> 
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cbc
> 
> 
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cbc