next up previous contents
Next: Data source Up: Project aims and motivations Previous: UDP vs TCP   Contents

Simulation systems

The simplest way of measuring the effect UDP data has on TCP is via simulation. Simulation provides both a way of controlling such parameters as connection bandwidth, router queue lengths and traffic levels, and ways to record such statistics as queue utilisation, packet loss and connection throughput. Attempting to obtain these sorts of results from an actual network would be both extremely costly and difficult.

Simulation however can only produce results as reliable and as accurate as the models employed. For simulation results to have any meaning, the component models in the virtual network, (such as switches, routers and most importantly, traffic models) must be as close to the real-world as possible. A traffic model provides a method for a simulator to create the data to flow through the network. Without accurate models to do this, the results of simulations would have no reference to real networks, creating pointless results.

This project luckily has access to a simulator of this quality. This is a discrete event simulator and is based on the ATM-TN simulator [ACGW95], with a range of modifications. The main change provides a more generic serial interface compared to the full ATM infrastructure present in ATM-TN. ATM-TN provides an actual TCP stack, derived from 4.4 BSD Lite, and a powerful simulation engine. More details including investigations completed using this can be found in [MPC98]

A traffic model for this simulator exists for HTTP traffic which makes use of TCP for transmission. These models will provide the reference traffic, the performance of which will be tested as UDP data rates are increased into the network.

Traffic models are based around network traces. The simulator takes data recorded from these network traces replicates data during simulation to provide loads for the network. If there is insufficient data to create the required loads the simulator offsets and overlays multiple copies of the recorded data.

For this project a set of traces is required to form a model for UDP traffic. The levels of the UDP traffic would then be increased over simulation runs, and the throughput achieved on the existing TCP traffic models can be measured.


next up previous contents
Next: Data source Up: Project aims and motivations Previous: UDP vs TCP   Contents
James Curtis 2000-01-17