Cost-optimal control policies for N-echelon divergent networks