The Design and Simulation of a Transport Protocol for Interactive Network Applications

The Internet is currently an IP datagram network, which uses the Transmission Control Protocol (TCP) for guaranteed data delivery. In addition to providing a reliable data transport layer, TCP aids the stability of a large scale internetwork through congestion detection and avoidance algorithms. Since TCP's inception in 1981, both the Internet and the applications which use it, have evolved. The result is a broad spectrum of data traffic, being transported by protocols which were developed twenty or more years previously. Increasingly, the traffic being carried over the Internet is part of an interactive client/server session, established between hosts on widely separated networks. The number of router hops between such hosts means that an aggressive transport protocol for application data may attempt to send data which exceeds the bottleneck capacity of a given network path. The result is packet loss which, for a guaranteed protocol, implies retransmission of data. Recent research has shown that currant implementations of TCP, based on the original TCP algorithms, are inappropriate for the transportation of modern Internet traffic. The thesis is concerned with the design, modelling, simulation and study, of an experimental transport protocol. We aim to address the issues face by current TCP implementations when transporting short, bursty, variable bit-rate, or bulk data in congested environments. In doing so, alternative methods of connection initialisation, flow control and congestion avoidance are implemented and simulated. Through simulation with bulk, variable bit-rate and live HTTP trace data, we show how our experimental protocol is able to deliver data with successful throughput comparable with currently implementations of TCP. Due to its modified startup and congestion avoidance algorithms, however, it does so with significantly reduced packet loss and overall packet transmissions.