SCASys: a smart congestion control algorithm selection system

Abstract

Although a lot of congestion control algorithms have been proposed in the past thirty years, researchers pointed out that there is no single one that can achieve best performance in all kinds of network environments. However, service providers mostly deploy one dedicated congestion control algorithm on their servers, which may result in some users not being able to get a high-quality experience. To address this issue, we propose a decision-tree based smart congestion control algorithm selection system named SCASys. SCASys models the link environment based on real-time statistical data, and periodically selects the most suitable congestion control algorithm in order to adapt to the dynamically changing link environment. We test SCASys in two types of environments: steady links and dynamic links. The result shows SCASys can have better environment adaptability and always achieve better performance in various scenarios compared with CUBIC and BBR.