Distributed Lyapunov drift-plus-penalty routing for WiFi mesh networks with adaptive penalty weight

In our previous work on distributed backpressure routing for WiFi mesh networks (WMN), we showed that backpressure routing when combined with geographic information results in a (quasi-)stateless low-overhead (hence scalable) routing solution. Specifically, the Lyapunov drift-plus-penalty optimization framework allows trading off between 1) routing decisions for maintaining queue backlogs under control (and hence, the network stable) and 2) those that try to get close to the optimal value of an objective performance metric. Such framework offers a non-negative parameter (V) for weighting both components. However, fixed-V policies cannot efficiently handle WMN traffic dynamics in practical setups, as they will lead to queue overflows and degradation of the objective metrics. In this paper, we propose the first practical distributed variable-V algorithm that takes routing decisions aiming at achieving ideal objective metric values, yet not incurring into queue overflows. Furthermore, we compare variable- and fixed- V policies under several network setups. The simulation results obtained with ns-3 show that the variable-V policy is not only able to obtain the same throughput as that of the best fixed- V policy, but also to remarkably reduce the variability and maximum values of packet delay.