Abstract: Congestion in network impacts the quality of service provisioning. And congestion control is an important of IP QoS. The traditional congestion control algorithms in router make drop decisions mainly according to the congestion status of local buffer resources independently. This multi-level independent congestion control causes the bandwidth-wasting problem （BW-CDF） when data flow congested. The BW-CDF problem is analyzed theoretically and a new congestion control algorithm （CC-AMR） is proposed, which is based on awareness of the congestion status of multi-level resources. The CC-AMR algorithm can synthetically utilizes the congestion status of resources in remote forward engines and their ports to manage the buffer of network processors, so that more reasonable congestion control decisions can be made. The CC-AMR algorithm has been implemented in the core router who adopts the switch-fabric and network processor based architecture successfully. To evaluate the performance improvement, the experiments are carried out to compare the performance of CC-AMR algorithm against SARED algorithm using the router. The experiment results show that the CC-AMR algorithm can enhance the total throughput of the router effectively during the periods of congestion compared with the SARED algorithm. For further details, the average throughput is increased by the CC-AMR algorithm for about 40.60% than the SARED algorithm under the same conditions.