Originally posted by: Roni.Mohammad

Dear all,

I am trying to figure out how to run a C++ project with cplex in a cluster size of 800 compute blades with:Two 12-core 2.5 GHz Intel Xeon (Haswell) processors per blade (24 cores/node and 19,200 cores total) . I will be using few of the nodes . I am trying to solve a large scale mixed integer problem.  

 Is there any example which shows step by step  procedure that demonstrates how to configure and  run a C++ project with cplex in a cluster?  

Any advice or direction would be helpful

Thanks in advance.

Roni

Originally posted by: T_O

From our experience, distributed MIP does not really help when you already have 24 cores per node. Of course, this might be very dependent on the actual problem instance, but do not expect too much. It might as well be faster when you use fewer cores. I would just use one node of the cluster, not more.

Best regards,
Thomas

Originally posted by: Roni.Mohammad

Thanks for your reply. I am new to experimenting cplex in cluster . I am trying to configure my problem to the cluster  

• Did you follow this process  http://www.ibm.com/support/knowledgecenter/SSSA5P_12.6.1/ilog.odms.cplex.help/CPLEX/UsrMan/topics/parallel_optim/distribMIP/11_cpp_eg.html?  
• I am was wallking through this tutorial 

  #include <ilcplex/ilocplex.h>
  #include <cstring>

  ILOSTLBEGIN

  static void
  usage (char const *program)
  {
     cerr << "Usage: " << program << " <vmc> <model>" << endl
          << "   Solves a model specified by a model file using " << endl
          << "   distributed parallel MIP." << endl
          << "   Arguments:" << endl
          << "    <vmc>    The virtual machine configuration file" << endl
          << "             that describes the machine that can be" << endl
          << "             used for parallel distributed MIP." << endl
          << "    <model>  Model file with the model to be solved." << endl
          << "   Example:" << endl
          << "     " << program << " process.vmc model.lp" << endl;
  }

  int
  main (int argc, char **argv)
  {
     char const *vmconfig = NULL;

     // Check command line length (exactly two arguments are required).
     if ( argc != 3 ) {
         usage(argv[0]);
        return -1;
     }

     // Pick up VMC from command line.
     vmconfig = argv[1];

     // Solve the model.
     int exitcode = 0;
     IloEnv env;
     try {
        // Create and read the model.
        IloModel model(env);
        IloCplex cplex(model);
        cplex.importModel(model, argv[2]);

        // Load the virtual machine configuration.
        // This will force solve() to use parallel distributed MIP.
        cplex.readVMConfig(vmconfig);

        // Solve the problem and display some results.
        if ( cplex.solve() )
           env.out() << "Solution value  = " << cplex.getObjValue() << endl;
        else
           env.out() << "No solution" << endl;
        env.out() << "Solution status = " << cplex.getStatus() << endl;

        // Cleanup.
        cplex.end();
        model.end();
     }
     catch (IloException& e) {
        cerr << "Concert exception caught: " << e << endl;
        exitcode = -1;
     }
     catch (...) {
        cerr << "Unknown exception caught" << endl;
        exitcode = -1;
     }

     env.end();

     return exitcode;

  }  // END main

  What are the three arguments require to run this program?
• Did you install cplex in cluster and submit the project/job to the cluster?

Thanks again for your feedback

Originally posted by: T_O

I assume, you have OpenMPI available on your cluster. Then please have a look at the attached script. Just call it as follows:

mpirun --npernode 1 ./cplex_mpi /path/that/contains/the/cplex/binary /path/to/your/problem/file/e.g./air04.mps.gz

This should do everything for you. Please obtain the nodes of the cluster completely and exclusively not to disturb other users.

Of course, you can do all this from inside a job script or an interactive job.

Best regards,
Thomas

If you never used MPI before then I suggest you first make yourself familiar with that software. It is completely unrelated to CPLEX but you will probably have a hard time to run CPLEX distributed parallel MIP through MPI if you don't know it.

The program requires only two, not three arguments. What these arguments are is explained in the program's usage message (see the usage() function): the first argument is the VMC file, the second is the model to solve.

You may want to read the distributed parallel MIP documentation to understand how things work and in particular what a VMC file is. Pay particular attention to the chapters that illustrate how to use either openMPI or MPICH and also to the chapter that explains how to use distributed parallel MIP in the interactive. Running your own code is very similar, you just have to run a different master program. If you only want to solve a problem then you don't even need custom code, you can just follow the instructions for running distributed parallel MIP in the interactive. Maybe it is best to start with that.