FAST TCP Congestion Control Module Implementation and Evaluation in Linux Kernel 2.6


Jianming Luo

Oral Defence Date: 

Friday, June 1, 2007 - 15:00


TH 935


3:00 PM


Professors Marguerite Murphy, Professor Dragutin Petkovic


The existing TCP congestion control algorithms (e.g. Reno, Vegas, Tahoe) are able to perform well in today’s low speed 10/100M Internet environments. It is also well known, however, that as bandwidth-delay products continue to grow, these approaches will eventually become a performance bottleneck themselves. FAST TCP was developed at the California Institute of Technology by Steven Low and Cheng Jin to address this issue for high-speed long latency networks. FAST TCP adjusts its congestion window size depending on the distance from equilibrium, which is defined as the optimum utilization of network capacity. This improves the speed of convergence and the protocol stability. As of Linux 2.6.13, new congestion control algorithms can be added as external kernel modules. The goal of the project presented in this report is to implement a FAST TCP module for the Linux 2.6 kernel for research and evaluation purposes, and then validate the implementation in an existing Internet environment as well as a simple local network environment. Subsequent experiments of this FAST TCP implementation show the congestion window size converges rapidly to a stable state and the protocol software performs relatively well in low speed environments. The contributions of this project include an implementation of the FAST TCP module in Linux 2.6, validation of the module in low speed networks, and a performance comparison with standard protocols.

Jianming Luo

TCP, FAST TCP, TCP TAHOE, TCP RENO, TCP VEGAS, High speed long latency networks, Equilibrium, Linux 2.6 Kernel, Congestion Window Size