| Abstract: |
Linux is getting to be a popular OS for both desktop and web-servers. Tuning can be done to increase the TCP/IP performance and maximize bandwidth. Obtaining good TCP throughput across a wide area network usually requires some tuning. This is especially true in current high-speed networks, where even though there is no congestion, an application may see only a small percentage of the total available bandwidth. This project includes simulating the lab-setup with an ns2 simulator with different TCP variants. Each TCP variant has its own congestion control algorithm. TCP Tahoe had the first congestion control algorithm. Since then a number of variants of TCP have been proposed: Reno, New Reno, Vegas, SACK, FACK, and Westwood. The current lab setup involves three computers. The goals of this project were to validate measured benchmark results of Web server workloads. Two computers are connected via Ethernet and third one is connected using 802.11g (Wireless). We have used this model to design the simulation and then run this simulation using different agents (TCP variants). Ns2 does not have built in support for creating sophisticated graphical presentation of collected data. We have written scripts to analyze the data and give appropriate conclusions. The simulation results show for wired connection (Ethernet) TCP Westwood had the best throughput. This is a relatively new TCP flavor with better congestion control algorithms. For wireless TCP, none of the TCP flavors had overall better performance metrics due to the high error rate of wireless media and consequent retransmissions.
|
