Sender-Based Approach to Reducing the Bufferbloat in Heterogeneous 5G Networks

The current 5G networks are demonstrably capable of delivering very high data rates. To maintain high data rates even under variable network conditions, they usually rely on very large network buffers that can store the data in poor channel conditions and output it again when conditions improve. In such cases, a TCP sender using loss-based congestion control will usually only experience increased latency. However, while the channel conditions are poor, the sender does not adjust its send rate and continues to fill up the buffer, which leads to bufferbloat.In this paper, we address the bufferbloat problem caused by loss-based TCP congestion control in networks with variable link capacities, especially in heterogeneous 5G environments. Unlike AQM where the mechanism must be deployed at the bottelneck link buffer, we propose a method that discards packets at the TCP sender when the RTT exceeds a predefined upper bound for a fixed time interval. This eliminates the need for changes by network providers. We developed our method by identifying optimal parameters using link emulation and validated its performance with both stationary and mobile measurements in a commercial 5G NSA-NR network in Hannover, Germany.