Improving Inter-protocol Friendliness Through Revision of Copa Competitive Mode and Estimation of BBR Behaviors Using Machine Learning

In the last decade, several influential congestion control algorithms (CCAs) have been devised for yielding high performance over the Internet. Among them, BBR, a CCA developed at Google in 2016, aims at attaining the optimum operating point reaching both the maximum throughput and minimum round-trip time simultaneously. Again, Copa, a delay-based congestion control assuming Markovian packet arrival has been developed at MIT and was first reported in 2018. Since BBR and Copa are likely to be among the major CCAs on the Internet in the coming days, it is important to ensure mutual friendliness between these two types of flows. However, that is often impossible due to the aggressive nature of BBR. To support scenarios where flows of different types share the same bottleneck link, Copa provides a specific operating mode, called competitive mode. However, even in competitive mode, while competing with BBR for network resources, Copa performance is adversely affected. In this paper, we therefore, propose a novel mechanism to improve the competitive mode of Copa. Also, using machine learning, we estimate the behavior of a BBR flow that competes for bandwidth with a Copa flow over the same path. Experiments using iPerf servers show that our proposed scheme improves the throughput of Copa by up to more than 700% when deployed together with BBR. This improvement leads our proposed scheme to achieve much higher friendliness with BBR than original Copa.