On the Use of TCP BBR in Cellular Networks

TCP BBR is a new TCP variant developed at Google, which, as of this year, is fully deployed in Google's internal WANs and used by services such as Google.com and YouTube. In contrast to other commonly used TCP variants, TCP BBR is not loss-based but model-based: It builds a model of the network path between communicating nodes in terms of bottleneck bandwidth and minimum round-trip delay and tries to operate at the point where all available bandwidth is used and the round-trip delay is at a minimum. Although TCP BBR has indeed resulted in lower latency and more efficient usage of bandwidth in fixed networks, its performance over cellular networks is less clear. This article studies TCP BBR in live mobile networks and through emulations, and compares its performance with TCP NewReno and TCP CUBIC, two of the most commonly used TCP variants. The results from these studies suggest that in most cases TCP BBR outperforms both TCP NewReno and TCP CUBIC. However, under precise network conditions, competing TCP BBR flows do not share the available bandwidth in a fair way, something that shows up, for example, when shorter TCP BBR flows struggle to get their fair share from longer ones.

[1]  Vern Paxson,et al.  TCP Congestion Control , 1999, RFC.

[2]  Andrea Francini,et al.  Link-Coupled TCP for 5G networks , 2017, 2017 IEEE/ACM 25th International Symposium on Quality of Service (IWQoS).

[3]  Martina Zitterbart,et al.  Experimental evaluation of BBR congestion control , 2017, 2017 IEEE 25th International Conference on Network Protocols (ICNP).

[4]  Marco Ajmone Marsan,et al.  Experience: An Open Platform for Experimentation with Commercial Mobile Broadband Networks , 2017, MobiCom.

[5]  Van Jacobson,et al.  BBR: Congestion-Based Congestion Control , 2016, ACM Queue.

[6]  Yuchung Cheng,et al.  BBR Congestion Control , 2017 .

[7]  Åke Arvidsson,et al.  TCP behaviour in LTE: Impact of flow start-up and mobility , 2016, 2016 9th IFIP Wireless and Mobile Networking Conference (WMNC).

[8]  Injong Rhee,et al.  Taming the elephants: New TCP slow start , 2011, Comput. Networks.

[9]  Sally Floyd,et al.  The NewReno Modification to TCP's Fast Recovery Algorithm , 2004, RFC.

[10]  Injong Rhee,et al.  CUBIC: a new TCP-friendly high-speed TCP variant , 2008, OPSR.