TCP Fairness In Ad Hoc Wireless Network

A host of problems are created by various Ad hoc wireless network characteristics not found in wired network, and not all of these problems have been satisfactorily solved. Therefore TCP performance in Ad hoc wireless network is not as stable as in wired networks. Recent work on TCP performance in Ad hoc wireless network mostly focuses on TCP efficiency (e.g. enhancing throughput), but we focus on TCP fairness. Significant TCP unfairness, even channel capture, has been observed through simulations. Fairness definitions and fairness indexs developed for the wired network cannot be directly applied to Ad hoc wireless network. Therefore we propose a Neighborhood MaxMin Fairness definition and it corresponding fairness index for evaluating TCP fairness in Ad hoc wireless network. Two factors, spatial reuse constraint and location dependency, are the most important factors, which affect TCP fairness in Ad hoc wireless network. Therefore, we take them into considerations in the definitions above. To improve TCP fairness, we analyze TCP unfairness from the transport layer view. A network layer solution called Neighborhood Random Early Detection (NRED) is then proposed. In NRED, we introduced the concept of neighborhood, which is the aggregate of a node itself and its interfering nodes, and the concept of distributed neighborhood queue, which is the aggregate of local queue at these nodes. We extend the RED concept to this distributed neighborhood queue. By detecting early congestion, notifying neighborhood nodes about congestion, and dropping packets proportionally to a flow's usage of channel bandwidth, the NRED scheme can improve TCP unfairness significantly. NRED is located at the network layer, and has no need of any modification in MAC layer. Simulation studies confirm that the NRED scheme can improve TCP unfairness substantially in Ad hoc network at the experse of throughput, but we think the experse is necessary.