DiffServ over MPLS networks is a widely accepted approach to considerably improve networks ability to support delay-sensitive applications such as voice over IP. In such networks, over-provisioning and careful admission control techniques are still needed, although insufficient to eliminate completely the congestion. To prevent congested paths, these networks use, currently, the preemption policy that induces waste of resources and excessive rerouting, which are triggered only after packets dropping. Besides, the increased end-to-end loss rates and delays, experienced by a service, are mostly due to one or few congested switches along the Label Switched Path “LSP” while the other routers are in relaxed conditions. In this paper, we tackle these issues by extending the traditional local management of congestions (isolated Active Queue Management “AQM”) into a cooperative process involving all switches along the service path at network operators scale. Going from AQM limitation, we propose, a network self-managing framework that dynamically re-adjusts switches parameters throughout the LSP. In this way, most of the prospective congestions impact is absorbed through balancing the routers aggressiveness without reconsidering other traffic engineering strategies (e.g., rerouting decision). Otherwise, the proposed framework allows rerouting at the point where congestion would most likely occur, which permits minimizing both packets loss and excessive rerouting. While considering QoS guarantees along a given service path, network loss ratio is reduced. This obviously allows network operators to further exploit theirs underlying resources by accepting more QoS-enabled services.