Besides the straightforward search for a single solution that you saw in the preceding section, it is also possible to implement a multiphase search in Parallel Solver, where each phase searches differently. To enable you to implement a multiphase search, Parallel Solver offers classes of such multithread objects as mutexes in IloFastMutex, semaphores in IloSemaphore, conditions in IloCondition, and barriers in IloBarrier so that your application can synchronize work more precisely among its workers.

The following example shows you how to use those multithread objects to implement multiphase search in your parallel application.

The complete program follows. You can also view it online in the file YourSolverHome/examples/src/whouse_mt2.cpp.

#if defined(ILCUSEMT) || defined(ILOUSEMT)
 
# include <ilsolver/ilopsolver.h>
ILOSTLBEGIN
 
void readCosts(const char*   name,
               IloInt&       buildingCost,
               IloIntArray2& costs,
               IloInt&       nbClients,
               IloInt&       nbWhouses) {
  ifstream in(name);
  in >> buildingCost >> costs;
  nbClients = costs.getSize();
  if ( nbClients ) nbWhouses = costs[0].getSize();
  else             nbWhouses = 0;
}
 
int
main(int argc, char** argv)
{
  IloInitMT();
  IloEnv env;
  try {
    IloInt i;
    IloModel model(env);
 
    IloInt       buildingCost, nbClients, nbWhouses;
    IloIntArray2 costs(env);
    const char* fileName;
    if ( argc < 2 ) {
      env.warning() << "usage: " << argv[0] << " <filename>" << endl;
      env.warning() << "Using default file" << endl;
      fileName = "../../../examples/data/whouse.dat";
    } else fileName = argv[1];
    readCosts(fileName, buildingCost, costs, nbClients, nbWhouses);
 
 
    IloIntVarArray offer(env, nbClients, 0, nbWhouses-1);
    IloIntVarArray transCost(env, nbClients, 0, 10000);
    IloBoolVarArray open(env, nbWhouses);
 
    for (i=0; i < nbClients; i++){
      model.add(transCost[i] == costs[i](offer[i]));
      model.add(open(offer[i]) == 1);
    }
 
    IloIntVar cost(env, 0, 10000, "Cost\t");
    model.add(cost == IloSum(transCost) + IloSum(open)*buildingCost);
 
    IloGoal goal = IloGenerate(env, offer) && IloInstantiate(env, cost);
 
 
    // end of model
 
    // We create the Parallel Solver
 
    IloParallelSolver psolver(model, 3);
 
    // We do a first run to tighten the problem
    psolver.solve(goal);
    IloSolver solver = psolver.getWorker(psolver.getSuccessfulWorkerId());
    IloInt maxCost = (IloInt)solver.getValue(cost);
    solver.out() << "Setting max cost to " << maxCost << endl;
 
    model.add(cost <= maxCost);
 
    // We search for an optimal solution
    model.add(IloMinimize(env, cost));
    psolver.solve(goal);
 
    solver = psolver.getWorker(psolver.getSuccessfulWorkerId());
    solver.out() << endl << "Optimal Solution" << endl;
    solver.out() << "Cost " << solver.getValue(cost) << endl;
    solver.out() << "Offer ";
    for (i=0; i<nbClients; i++)
      solver.out() << solver.getValue(offer[i]) << " ";
    solver.out() << endl;
    solver.out() << "TransCost ";
    for (i=0; i<nbClients; i++)
      solver.out() << solver.getValue(transCost[i]) << " ";
    solver.out() << endl;
    solver.out() << "Open ";
    for (i=0; i<nbWhouses; i++)
      solver.out() << solver.getValue(open[i]) << " ";
    solver.out() << endl;
    psolver.end();
  }
  catch (IloException& ex) {
    cout << "Error: " << ex << endl;
  }
  env.end();
  IloEndMT();
  return 0;
}