I have asked this question before as "increasing solution time in a loop" here; https://www.ibm.com/developerworks/community/forums/html/topic?id=0ef4c99b-3bcc-4105-ac98-57867d5427da but the forum is currently not available.
Later, I realized that quadratic, second order cone programming constraints create a memory leak. Here is a summarized version of the code (source and the header can be found in the forum link above):
#include <header.h>
int main() {
const int n = 10;
const int ins = 30;
// Parameters are read
for (int i = 0; i < ins; i++) {
IloEnv env;
IloModel model(env);
IloNumVarArray d(env, n + 1, 0, IloInfinity, ILOFLOAT);
IloNumVarArray A(env, n + 1, -IloInfinity, IloInfinity, ILOFLOAT);
IloNumVarArray C(env, n + 1, -IloInfinity, IloInfinity, ILOFLOAT);
// Other variables are defined
model.add((d[1] * d[1]) >= (A[1] * A[1]) + (C[1] * C[1])); // SOCP constraint 1
model.add((d[2] * d[2]) >= (A[2] * A[2]) + (C[2] * C[2])); // SOCP constraint 2
// Linear constraints are added according to the parameters of the instance
// Solve method of IloCplex is called and outputs are collected
env.end();
}
}
When this code is executed, solution times are increasing with an increase in i and memory usage tool of Visual Studio shows that there is a memory leak with each loop. Memory leak disappears if SOCP constraints are commented out. Strangely, if only one of those SOCP constraints is added, either 1 or 2, memory does not leak. It leaks only if both of them (or more other SOCP constraints) are added at the same time.
Is there a problem of my way to add these quadratic constraints?
Thank you.
Edit: To reproduce file, you can use the following link: https://drive.google.com/open?id=1AAPaSUhGwdGZ15c3MEb0atRQsrAel-mJ. It includes the header, the source and the inputs.
