ETX Could Result in Lower Throughput

ETX (Expected Transmission Count) is the best routing metric for single radio wireless mesh network (WMN) in terms of throughput as it selects the links with highest packet delivery paths. In our experiments, however, we found that under some scenarios which commonly occur in normal practice, e.g., when data rates approach channel capacity, ETX could yield lower throughput than other metrics, such as, minimum hop-count or RTT (Round Trip Time). The paper presents the experimental evidence of performance anomaly in ETX, identifies the operating conditions when anomaly occurs, and discusses the results of our investigation regarding potential causes, such as, probing overhead, load-sensitivity, etc., using simulations and testbed experiments. Results show that probing, though incurs small overhead, plays a major role in performance degradation of ETX. When node buffers are full because of higher data rate, prioritized control packets, including probes, push data packets outside the queue. When alternative paths are not very much different to each other, ETX appears worse than hop-count because of this additional buffer overflow loss. Although, this loss becomes insignificant when penalty of selecting a lossy path is higher. We also observed that ETX exhibits route oscillations due to load-sensitivity but strong correlation of its load-sensitivity with degraded performance is not found.