Mortise: Auto-tuning Congestion Control to Optimize QoE via Network-Aware Parameter Optimization

Mortise

Abstract

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.

Publication
To appear in the 23rd USENIX Symposium on Networked Systems Design and Implementation 2026
Jing Chen
Jing Chen
Ph.D. of Computer Networking

My research interests include low-latency network transport, interactive video streaming and wireless networks.