Measurement-Based Short-Term Performance Prediction in Wireless Mesh Networks

Traditionally, the performance of wireless mesh networks (WMNs) is measured by long-term averaged metrics, such as long-term averaged packet delivery ratio or throughput. However, due to the dynamic nature of the wireless networks, long-term averaged metrics cannot reflect the short-term behaviors of the network. In the meanwhile, the users may require a sustained performance for a certain period of time in many realtime applications. Prediction of network performance of WMNs at the level of individual flows in a small time granularity becomes very important, but is missing in the literature. In this paper, we propose a measurement-based model to predict the short-term performance of both goodput and packet loss for individual flows in WMNs. This model captures the complex dependencies among the different queues in the system, traffic demand, and wireless interference in the network. We developed two tools Rater and CalMedium to calculate the probabilities of packet loss along all the layers in the protocol stack at each node. Real-world experiments on an indoor mesh network testbed demonstrate that our prediction method can achieve accurate performance prediction under both single-flow and multiple-flow scenarios. We also discuss two application examples of utilizing our prediction model: finding the bottleneck rate of a flow and admission control.