An Approximating Fair Congestion Control Scheme at Core Network

Congestion is a natural phenomenon in core network, and is unavoidable if the queuing system is operated near capacity. Providing fairness among users with different levels of aggressiveness, and different levels of responsiveness is necessary for efficient network flow control. The scalability of current solutions depends on the number of active flows and the number of flows whose rate is greater that the current fair share. In this paper we propose a new fair active queue management scheme, called SDP (selective differentiated penalization), that use the recent buffer resources allocation history to differentially penalize misbehaving flows by dropping more of their packets to make a tradeoff between scalability and fairness. The first principle of SDP is that the resulting throughput of each flow should be bounded by the max-min fair share bandwidth. The performance of SDP is aided by the fact that the recent history of buffer resource allocations can provide accurate rate estimates of active flow and the fair share threshold. Simulated results suggest that SDP provides a reasonable degree of max-min fair bandwidth allocation in a wide variety of operating conditions.