Ladygina O. Metods of Statistical Optimization of Servicing Unstable Traffic in Heterogeneous Computer Networks. Manuscript

The objective of the dissertation is to increase the efficiency of non-stationary traffic service in heterogeneous computer networks by statistical optimization of non-stationary traffic service taking into account the existing limitations and improvement of non-stationary traffic models to increase their adequacy to real conditions. The tasks to build the improved models of dynamics of polynomial and periodic traffic were set and solved; traffic periodicities on the background of obstacles were disclosed; the structures and parameters of traffic which has random periodic character were determined, that is the numbers and parameters of harmonics in the structure of unstable traffic were also determined; the methods and algorithms of statistical optimization of servicing traffic in simplex, duplex, many-server systems and heterogeneous computer networks were developed; mathematical models of the criteria of optimality of servicing traffic and limits for heterogeneous and computer networks were developed. Every problem description presents grounding of the choice of necessary data, method of task solving and expected results. The statistical data of the direct measuring of traffic in heterogeneous computer networks are used as incoming data and the results of the observations of their operation. The grounding of the influence of the choice of number of quantization levels on the accuracy indicators in constructing models by quantization method, Markov approximation and model improvement parameters was presented. The dependence of optimality criteria on random parameters, the choice of method for optimization and search for optimal solving were given. The analysis of dependences of optimal decision on the parameters of functioning of many-server system of servicing traffic was also offered. The task of verification of the improved models of polynomial unstable traffic and the structure and parameters of multi-component periodic traffic on the background of obstacle was solved.