Congestion control in long fat networks

Andrej Mohar (2012) Congestion control in long fat networks. EngD thesis.

Preview

PDF Download (2275Kb)

Abstract

In the present time the TCP protocol is used widely for the data transmission. People are constantly trying to speed up the data transmission over the Internet. The networking speed is not dependent on the fast Internet networks alone, but on the connection quality as well. The first big congestion collapse happened in the year 1969 on the ARPANET network. After that event the new algorithm was added to the base TCP protocol, which slows down the sending rate thus stops congestion from happening. From that point on, the new TCP variants are being developed, which try to predict the right sending speed so that the network will be fully utilized without major congestion collapses. The thesis describes the basic TCP protocol elements, which participate in the congestion avoidance process. It introduces the modern fast networks and the problems which arise during the communication on such fast networks. The different simulators and simulations are enumerated and described in the last part of the thesis. The simulation results describe the similarities and differences on several different graphs: congestion window, the queue usage, the queue delays, dropped packets, round trip times and bandwidth usage. The simulations are used on a simple model containing the “chokepoint” router. The following protocols are simulated: TCP Reno, TCP Vegas, TCP Cubic, TCP Hybla and TCP Compound. The TCP Reno has the lowest number of sent and received packets. The results show the TCP Vegas as the best algorithm, but no other flows are present in the simulation. The second best is the TCP Cubic protocol, which can be used on both wired and wireless networks.

Actions (login required)