Originally posted by: lmalbino

Hello, i'm trying to run a model(39054 Variables and 2159478 constraints) from a lp file with importModel function (c++ IloCplex), and the function takes about 30000 seconds to read the model. If I try to read the model from the cplex console with the read command, the model is loaded in 7 seconds. Any justification for this difference? I've seen some earlyer topic about this function and what I understand, was that in version 12.5.* exists some problem and that version 12.6 already solve that problem. Probabily I don't understand the user or the problem was different.
Any help about his issue?
Thank you for the attention.

 Luis Albino

Does your model include any quadratic terms, either in the objective or in constraints?

Does it help to call IloEnv::setNormalizer(false) before calling IloCplex::importModel()?

Could you post your model here?

Originally posted by: lmalbino

Hello,

No quadratic terms.

Doesn't help, I already have tried that with version 12.5.1, and it was in the implemented code. So the loading times that I present are with that parameter defined.

I' ve upload the model file.

Thank you for the help

Best regards

Thanks for the file.

I could reproduce the problem here. I have not figured out yet what the problem is but I found a simple workaround: Use much shorter names. When I rename the variables to x1 through xN and the constraints to c1 and cN then importModel() runs much faster. Is it an option to use shorter names? If so, then you should not create an .lp file but an .rlp file. That will store those generic names instead of the original names. Or even better, export as a .sav file without any names.

We will try to figure out why import with longer names is so slow.

Originally posted by: lmalbino

Hello, thank you for your help.

For now the problem is solved, we have reading times of 9 seconds with .rlp files.

Thank you

Best Regards

We have found (and fixed) what causes this performance problem. A workaround that allows you to keep the long names is to replace code like

IloModel model(env);
IloCplex cplex(model);
cplex.importModel(model, ....);

by

IloModel model(env);
IloCplex cplex(env);
cplex.importModel(model, ...);
cplex.extract(model);

The point here is that at the time you call importModel() cplex does not yet have the model extracted. You could also do

IloModel model = cplex.getModel();
cplex.clear();
cplex.importModel(model, ...);
cplex.extract(model);

With any of the two rearranged code, reading the file should be almost as fast as in the interactive.

Originally posted by: Nadere

Hello,

I have SOCP model, when the program reaches the line "cplex.extract(model)",  it takes several minutes to pass it. here are some constraints in the model:

linear constraints;

    for (p = 0; p < m; p++)
                for (d = 0; d < n; d++)
                    for (i = 0; i < o; i++)
                        for (j = 0; j < adj_list[i].size(); j++)
                            model.add(ca1[p][d][i][j] * ca1[p][d][i][j] + ca2[p][d][i][j] * ca2[p][d][i][j] <= ca3[p][d][i][j] * ca3[p][d][i][j]);

            for (p = 0; p < m; p++)
                for (d = 0; d < n; d++)
                    for (i = 0; i < o; i++)
                        for (j = 0; j < adj_list[i].size(); j++)
                            model.add(ca4[p][d][i][j] * ca4[p][d][i][j] + ca5[p][d][i][j] * ca5[p][d][i][j] <= ca6[p][d][i][j] * ca6[p][d][i][j]);
            for (p = 0; p < m; p++)
                for (d = 0; d < n; d++)
                    for (i = 0; i < o; i++)
                        for (j = 0; j < adj_list[i].size(); j++)
                            model.add(ca7[p][d][i][j] * ca7[p][d][i][j] + ca8[p][d][i][j] * ca8[p][d][i][j] <= ca9[p][d][i][j] * ca9[p][d][i][j]);
               for (p = 0; p < m; p++)
                    for (d = 0; d < n; d++)
                         for (i = 0; i < o; i++)
                        for (j = 0; j < adj_list[i].size(); j++)

               model.add(ca10[p][d][i][j] * ca10[p][d][i][j] + ca11[p][d][i][j] * ca11[p][d][i][j] <= ca12[p][d][i][j] * ca12[p][d][i][j]);
                    
 model.add(IloMaximize(env, obj)); // linear objective function

obj.end();

IloCplex cplex(env);
cplex.extract(model); //*************************//

 cplex.setOut(env.getNullStream());

 cplex.setParam(IloCplex::Param::Barrier::QCPConvergeTol, 0.03);
 cplex.solve();

My question that is there any way to speed up extracting the SOCP model? 

Thanks,

Nadere 

I think you created a new thread and asked the exact same question here: https://www.ibm.com/developerworks/community/forums/html/topic?id=49e4d06b-4851-493b-a5b8-ce0e157c98e4