Model-driven optimization of multihop wireless networks

Interference is fundamental to wireless networks. It is hard to achieve good performance when design routing metrics or algorithms without taking it into account. We study interference in wireless networks through empirical experiments and simulations. We find out that current routing protocols face difficulties in effectively managing it, which can lead to severe problems. For instance, a simple network of two links with one flow is vulnerable to severe performance degradation if interference is not properly accounted for. Motivated by these observations, we develop a simple and effective model to capture effects of interference in a wireless network. Different from the existing interference models, our model captures IEEE 802.11 DCF under both homogeneous and heterogeneous traffic and link characteristics, and is simple enough to be directly used as a basic building block for wireless performance optimization. Based on this model, we develop optimization algorithms for several objectives, such as network throughput and fairness. Given traffic demands as input, these algorithms compute rates at which individual flows must send to achieve these objectives. We implement these algorithms in Qualnet simulations and 19-node testbed. Our experiment and simulation results show that our methods can systematically account for and control interference to achieve good performance. More specifically, when optimizing fairness, our methods can achieve almost perfect fairness; when optimizing network throughput, they can lead to 100-200% improvement for UDP traffic and 10-50% for TCP traffic.