A hierarchical scheme is proposed for the optimal control of large-scale non-linear systems with multiple state and control delays. A linear approximation of a nonlinear system is defined about the precomputed and stored nominal trajectories. The system is then decomposed into N subsystems such that the delay terms and interaction terms can be treated as an extra input, named perturbation, to the linear non-delay subsystems. The scheme generates a partly closed-loop and partly open-loop control. The convergence of the algorithm is established. An example with quadratic cost is simulated and solved by the proposed scheme using the constant nominal trajectories. The results of computer simulation of the two control problems presented in the examples are included.