Core-stateless fair bandwidth allocation for TCP flows

The standard end-to-end flow control implemented by the TCP protocol is ill-suited when it comes to achieving fair bandwidth allocation among competing TCP flows. Indeed, the lack of feedback from intermediate nodes does not allow a TCP source to regulate its throughput so that it is not sending more than its fair share, thus penalizing other, less aggressive flows. We propose a novel scheme that, building on existing work on network-layer stateless fair queueing, extends the approach to the TCP layer. Also, we discuss a possible implementation of both the network-layer and the transport-layer architecture. We test our solution under different traffic scenarios and show that not only is a fair bandwidth allocation achieved, but the overall network utilization for TCP flows is also increased.