On the probabilistic complexity of finding an approximate solution for linear programming