This paper proposes a scalable algorithm called APD (Adaptive Packet Dropping) for bandwidth control, which is based on active queue. A flow is mapped to an entry of hash table by hash function, and the counter of a flow can be searched at O(1) time. Theoretical analysis and experiments indicate that APD algorithm has great advantages on speed and space and it can reach predictable accuracy on flow bandwidth control. Theoretical analysis is also used to predict the accuracy for miscellaneous data-flow sizes distribution. A bandwidth control system is realized basedon Linux 2.6 kernel. The experimental results show that the system can work properly on gigabit links while keeping enough accuracy on flow based bandwidth control.

This paper proposes a scalable algorithm called APD (Adaptive Packet Dropping) for bandwidth control, which is based on active queue. A flow is mapped to an entry of hash table by hash function, and the counter of a flow can be searched at O(1) time. Theoretical analysis and experiments indicate that APD algorithm has great advantages on speed and space and it can reach predictable accuracy on flow bandwidth control. Theoretical analysis is also used to predict the accuracy for miscellaneous data-flow sizes distribution. A bandwidth control system is realized basedon Linux 2.6 kernel. The experimental results show that the system can work properly on gigabit links while keeping enough accuracy on flow based bandwidth control.