On the Incompatibility of Scalable Congestion Controls over the Internet

In addition to classic ones, a new family of congestion controls regarded as “scalable” has recently emerged. Scalable congestion control exploits ECN to provide much finer control as rate scales, enabling reactions proportional to the congestion level. Nowadays, different initiatives propose the use of Scalable congestion control over the public Internet, with the separation of Scalable and Classic flows. Similar to classic approaches, the evolution of Scalable congestion control has also begun as BBRv2 implemented a “DCTCP-inspired” ECN mechanism. In this paper, we investigate the compatibility of two available Scalable congestion controls, Data Center TCP, and Google' s recent proposal called BBRv2, and evaluate them with different ECN marking strategies. Our evaluation relies on numerous measurements carried out in our testbed with 1 to 10Gbps bottleneck capacities and heterogeneous round trip times. At the bottleneck, the ECN marking of two traditional AQMs (STEP and PI2) most commonly proposed for Scalable flows and a non-traditional core-stateless AQM (CSAQM) using in-network resource sharing control is examined. After showing compatibility issues, we conclude that as long as the end-to-end flows control resource sharing, it will be tough to create a new, evolved congestion control that is fair to legacy solutions, and its deployment does not harm the Internet.