v m rclIérenr, Our L'auteur conserve la propriété du droit d'auteur qui protège cette thèse. Ni la thèse ni des extraits substantiels de celle-ci ne doivent être imprimes ou autrement reproduits sans son autorisation. Abstract The ability to estimate performance parameters and congestion rates in high speed networks is a feature that is vecy useful to network service providen. [t lets them anticipate congestion problems brfore they occur. allowing them to avoid the problem. This thesis provides

more »

... empirical cvidence that supports a method that uses the past traffic characteristics of a network to predict its future performance. The data that is analyzed in this project cornes €rom an existing telephone Company ATM network. The Q value. which is the ratio of the mean traffic on a device to its standard deviation is used to classify the devices' traffic distribution. Evidence is presented that shows that on the studied ATM network. trafic distributions are relatively stable over time and that a devices' Q value can be used to characterize its traffic distribution. The relationship between the Q value and both the sampling interval and the sarnpling length is explored. This thesis dso shows that for self sirnilar data, the Q value can be estimated for a different sampling interval than the i nterval actuaily measured. Problems associated with sirnulating a large, high speed ATM network are presented in this rhesis. A method that takes a simulation generating a large number of cells per second and scales the simulation down to run in a reasonable time frarne is outlined. Acknowledgments 1 wouid like to thank Professor Nick Dawes and Professor Bernie Pagurek for their supervision and guidance throughout the work of this thesis. The motivation and direction they provided was greatly appreciated-I would also like to thank Tom Kanary. from Bell Sygma, for the time he invested in this project and for the ATM data that he provided.