Uploaded June 2026 | Updated September 2026, 3 weeks ago
Mortise: Auto-tuning Congestion Control to Optimize QoE via Network-Aware Parameter Optimization
Yixin Shen, Tsinghua University, Bytedance Inc., and Zhongguancun Laboratory; Ruihua Chen, Tsinghua University; Bo Wang, Tsinghua University and Zhongguancun Laboratory; Jing Chen, Haochen Zhang, and Minhu Wang, Tsinghua University; Yan Liu, Bytedance Inc.; Mingwei Xu, Tsinghua University and Zhongguancun Laboratory; Zili Meng, Hong Kong University of Science and Technology
Congestion control algorithms (CCAs) critically shape the tradeoff among throughput, latency, and loss, directly impacting user Quality of Experience (QoE).However, most existing CCAs use static, heuristically chosen parameter settings that fail to adapt to dynamic network states, resulting in suboptimal QoE. Our key observation is that the optimal CCA parameter configuration depends on real-time network states.To bridge this gap, we propose Mortise, a real-time, network-aware adaptation framework that dynamically tunes rule-based CCA parameters to maximize QoE. To address the challenges in modeling the complex parameter-QoE relationship, Mortise introduces a QoS tradeoff proxy to decompose parameter optimization into two steps: it first infers the application's preferred QoS tradeoff from real-time QoE gradients and then derives the corresponding parameter settings via control-theoretic analysis.Implemented atop TCP and evaluated in both emulated and production environments, Mortise outperforms state-of-the-art solutions, enhancing the QoE of file downloading service by up to 73% and QoE of video streaming service by up to 167% in real-world scenarios, with minimal deployment overhead.
View the full NSDI '26 program at usenix.org/conference/nsdi26/technical-sessions
Mortise: Auto-tuning Congestion Control to Optimize QoE via Network-Aware Parameter Optimization
Yixin Shen, Tsinghua University, Bytedance Inc., and Zhongguancun Laboratory; Ruihua Chen, Tsinghua University; Bo Wang, Tsinghua University and Zhongguancun Laboratory; Jing Chen, Haochen Zhang, and Minhu Wang, Tsinghua University; Yan Liu, Bytedance Inc.; Mingwei Xu, Tsinghua University and Zhongguancun Laboratory; Zili Meng, Hong Kong University of Science and Technology
Congestion control algorithms (CCAs) critically shape the tradeoff among throughput, latency, and loss, directly impacting user Quality of Experience (QoE).However, most existing CCAs use static, heuristically chosen parameter settings that fail to adapt to dynamic network states, resulting in suboptimal QoE. Our key observation is that the optimal CCA parameter configuration depends on real-time network states.To bridge this gap, we propose Mortise, a real-time, network-aware adaptation framework that dynamically tunes rule-based CCA parameters to maximize QoE. To address the challenges in modeling the complex parameter-QoE relationship, Mortise introduces a QoS tradeoff proxy to decompose parameter optimization into two steps: it first infers the application's preferred QoS tradeoff from real-time QoE gradients and then derives the corresponding parameter settings via control-theoretic analysis.Implemented atop TCP and evaluated in both emulated and production environments, Mortise outperforms state-of-the-art solutions, enhancing the QoE of file downloading service by up to 73% and QoE of video streaming service by up to 167% in real-world scenarios, with minimal deployment overhead.
View the full NSDI '26 program at usenix.org/conference/nsdi26/technical-sessions
