SmartSBD: Smart shared bottleneck detection for efficient multipath congestion control over heterogeneous networks
Enhuan Dong,Peng Gao,Yuan Yang,Mingwei Xu,Xiaoming Fu,Jiahai Yang
DOI: https://doi.org/10.1016/j.comnet.2023.110047
IF: 5.493
2023-10-02
Computer Networks
Abstract:Multipath TCP (MPTCP) has been widely adopted in today's mobile devices. However, two types of congestion control algorithms, uncoupled congestion control ( Uncoupled CC ) and coupled congestion control ( Coupled CC ), cannot achieve both bottleneck friendliness and throughput maximization for both of the MPTCP subflow bottleneck sharing scenarios, shared bottleneck ( SB ) scenario and non-shared bottleneck ( NSB ) scenario, leading to performance degradation in practice. In this work, we seek to enable efficient MPTCP congestion control, by alternating between Uncoupled CC algorithms and Coupled CC algorithms via smartly detecting whether the two MPTCP subflows share the same bottleneck link. We propose SmartSBD , the first learning-based data-driven approach for shared bottleneck detection, which is accurate, adaptable, and easy-to-deploy. SmartSBD is based on the key insight that the properties of subflows that share the same bottleneck often have similar trends of variation or similar values. In the training phase, SmartSBD collects system logs when MPTCP is running in real-world heterogeneous networks, extracts features, and trains a binary classifier. In the runtime phase, SmartSBD makes periodic predictions on the bottleneck sharing condition of live MPTCP subflows, and uses the prediction results to alternate between Coupled CC and Uncoupled CC . Our evaluations demonstrate that SmartSBD outperforms existing approaches.
computer science, information systems,telecommunications,engineering, electrical & electronic, hardware & architecture