This project explored the design, implementation, and evaluation of SUSS (Speeding Up Slow-Start), a lightweight enhancement to TCP’s initial congestion control phase. Slow-start traditionally increases the congestion window conservatively, often leading to bandwidth underutilization and prolonged flow completion time in high bandwidth–delay product networks. SUSS addresses this by introducing predictive growth based on ACK clocking stability and packet pacing, enabling faster ramp-up without requiring router modifications or changes to network infrastructure. Integrated into the Linux kernel’s CUBIC module, SUSS demonstrated significant improvements in high-speed environments while remaining fully backward-compatible with existing TCP stacks.
Under this project, I implemented SUSS within Linux kernel 5.19.10 and evaluated its performance through controlled testbeds and real-world cloud deployments, including Google Cloud and Oracle data centers. The work involved developing growth-factor prediction logic, integrating pacing mechanisms, and performing throughput, fairness, and FCT benchmarking. Results consistently showed 20–50% reductions in flow completion time and increased stability compared to CUBIC and BBR, particularly for short-lived flows that dominate modern web traffic. This project provided hands-on experience with congestion control algorithm design, kernel development, and empirical network research, demonstrating the practical potential of SUSS as a deployable enhancement to TCP in high-performance computing and data center environments.