Summer Research Internship at New York University under Prof. Shivendra Panwar and Prof. Fraida Fund
This repository contains the results of the experiments conducted during my on-site summer research internship at New York University. The project focused on analyzing the performance of various TCP congestion control algorithms and exploring the Low-Latency Low-Loss Scalable Throughput (L4S) framework.
- Analyze TCP Congestion Control Algorithms: Evaluate the performance of TCP Reno, TCP Cubic, and TCP Vegas, including their behavior in network scenarios.
- Investigate Coexistence of Algorithms: Study the performance when loss-based algorithms (like TCP Reno) and delay-based algorithms (like TCP Vegas) coexist in a network.
- Evaluate L4S Components: Examine how L4S components, including scalable sender and Enhanced ECN/AccECN, address problems faced in previous network scenarios.
- Study Dual Queue Coupled AQM: Understand how L4S traffic coexists with classic traffic using Dual Queue Coupled Active Queue Management (AQM).
-
TCP Congestion Control Experiments:
- Conducted experiments demonstrating AIMD (Additive Increase Multiplicative Decrease), slow start, and the classic sawtooth pattern of congestion window.
-
Performance Analysis:
- Analyzed TCP Reno, Cubic, and Vegas individually and in scenarios where they share the same bottleneck router with FIFO queue.
- Experimented with progressive network scenarios from classic TCP flow to Active Queue Management (AQM) flow (varying ECN threshold) and finally to L4S flow.
- I replicated the following 4 scenarios and analysed the performance and shortcomings of each scenario which led me to the next scenarios:
-
L4S Flow Analysis:
- Demonstrated the scalable sawtooth behavior of L4S and the operation of Enhanced Explicit Congestion Notification (ECN).
- Examined impacts on latency, bandwidth, and other parameters in networks coexisting of L4S traffic and classic traffic.