Self-similarity On Network Performance

In this paper, several mathematical definitions of self-similarity are given, and some of their properties are described. The method of justifying a self-similar process and getting its autocorrelation coefficient (or self-similar parameter) are also discussed. Achievements which are achieved by scientific research of predecessors are summarized. Some methods of how to generate self-similar process and a few means of estimating self-similar parameter are given. Self-similar traffics are generated by using of ON/OFF model with heavy-tailed distribution and results are given. The results of experiment accord with the theory. This identicalness indicates that the method of getting results by using of ON/OFF model is feasible.In network performance, queueing performance is an important indicator. Methods of queueing analysis in traditional theory are based on mature theory. Laplace transform is a crucial mathematical tool which is used by classical methods. It is the heavy-tailed distributions that may not possess all of their moments, and, therefore certainly not their Laplace transforms that brings the difficult to analyzing heavy-tailed distribution service time. In order to solve this problem, a new method is proposed in this paper.CLP (cell loss probability) is another main factor in network performance. A method of estimating the CLP in an ATM (asynchronous transfer mode) multiplexer which is fed by a self-similar arrival process is advanced in this paper. The approach is based on theory of large deviations, and the simulation tests show that it is more accurate than the existing approximation results. The analysis and results provide the theoretical basis of designing buffers and bandwidth.