We present a reformulation-based dual ascent approach for the Multi-Commodity network Flow problem. More specifically, we consider the path-flow formulation of the linear relaxation of the problem and we solve it with a column generation. We apply a suitable reformulation to each restricted master problem of the column generation in such a way as to obtain a Lagrangian relaxation that can be decomposed into smaller and easier sub-problems and we apply a dual ascent step. This allows to efficiently obtain a good quality dual bound of the linear relaxation of the problem and we show comparisons with other approaches.