HACC: Hierarchical Automatic Selection of Congestion Control Algorithms

Most congestion control mechanisms work well in certain network environments, and no one can adapt well to consistently deliver good performance in all scenarios. The recently proposed frameworks based on reinforcement learning can flexibility select congestion control algorithms to be resilient to the dynamic changes for network status. However, frequent alterations to the congestion control mechanisms during the relatively stable period of the network actually lead to network instability and unnecessary computational overhead. In this paper, we present a hierarchical adaptive congestion control algorithm (HACC) to solve this problem. HACC dynamically selects the appropriate congestion control mechanism only when the current congestion control algorithm is not suitable for the current network state, rather than changing congestion control scheme every training cycle to ensure network stability. The simulation results show that HACC reduces flow completion time (FCT) and increases throughput significantly in stable network status. For example, ACC achieves throughput rates that are 47%, 35%, 23% and 8% over Cubic, Reno, BBR and Antelope.