On active queue management in cellular networks

Active queue management (AQM) is broadly accepted as the tool of choice for tackling bufferbloat (the accumulation of large queueing delays) in network buffers. CoDel and PIE are considered two of the most promising AQM schemes, but little has been done to adopt them in cellular networks, which are particularly plagued by bufferbloat. This is partly due to the lack of adequate evaluation of their performance on links where the data rate varies quickly and widely. Past studies have used simplistic link models and thus do not provide a strong case for the adoption of either scheme in cellular networks. We study the performance of CoDel and PIE on a recently introduced emulator that accurately reproduces the highly variable nature of cellular links. We discover that both schemes incur link utilization losses up to 60% in return for reducing queueing delay. This trade-off is undesirable for network providers and applications, yet it is inherent to the design of both CoDel and PIE. We argue that AQM schemes designed around the primary goal of maximizing network power (the ratio between throughput and queuing delay) are not the most effective tool for tackling bufferbloat in cellular links. We propose a new queue balance metric as the driving criterion for designing more effective congestion control solutions for cellular networks.