Simulation And Experiment Of "Neighborhood RED" In Wireless Ad Hoc Network

Many problems are induced by various wireless ad hoc network characteristicsnot found in wired network and some of them have not been completely solved. SoTCP performance in wireless ad hoc network is not as stable as in wired network. Inorder to improve TCP fairness of wireless ad hoc network, Dr. Xu of UCLA proposeda network layer solution called Neighborhood Random Early Detection (NRED). InNRED, the concept of neighborhood is introduced, which is the aggregation of a nodeitself and its interfering nodes. So the RED mechanism is extended to the distributedneighborhood queue, which is the aggregation of local queue in one's neighborhood.However, Dr. Xu simulated and analyzed the performance of NRED by measuringchannel utilization rate, which is an over-layer solution and difficult to implement inpractice.To validate the effectiveness of NRED, we implemented it by broadcastingqueue length in network layer in the simulation and experiment. Both of them showedthat the performance of this scheme is not consistent with NRED theory. There aretwo main problems. First, broadcasting packets are not reliable;some of them arediscarded when the channel is congested. Second, congestion areas detected in theexperiment are not those given by theory, because the neighborhood queue length isnot accordant with channel utilization rate. So we draw the conclusion that the schemeof broadcasting queue length was not effective.To avoid the problems mentioned above, we proposed another two fairnesssolutions in network layer based on passive measurement. The first is theimprovement of NRED based on average packet delay measuring. In this scheme, wemeasure one-hop delay in multi-hop ad hoc network, using that to detect congestion.So we can control drop rate to guarantee fairness. The second is based on bandwidthutilization rate measuring, which is a new fairness solution. In this scheme, everynode monitors its throughput separately. If the bandwidth utilization rate exceeds thethreshold we set, the traffic of this node should be paused so that other nodes willhave chance to send packets. We implemented this scheme in test bed, and comparedthe experiment result with simulation finished by Lee Yeng-zhong of UCLA. Both ofthem accordantly prove that passive measurement is more precise and the schemebased on bandwidth utilization rate measuring can greatly improve TCP fairness.