The dissertation is devoted to the creation of methods and devices for the formation of flows of heterogeneous network traffic. It is shown that such heterogeneous traffic is self-similar (fractal). The main feature of self-similar traffic is the occurrence of rapid sporadic bursts of intensity at average and relatively low traffic intensity at long data transfer intervals. This leads to the growth of queues in the buffer memory and, as a result, overloads of switching nodes. Therefore, the task of researching and developing new methods of building traffic shaping devices with adaptation to changes in network parameters and status is urgent.
The aim of the dissertation is to increase the efficiency of the functioning of telecommunication networks of new generations by eliminating overloads with hardware and software means of adaptive transformation of incoming traffic statistics.
The following new scientific results were obtained in the dissertation work.
1. The model for managing the parameters of information flows in telecommunication networks has been improved. Unlike the existing models, the proposed model is built on the basis of the theory of Markov processes, which allows the analysis of self-similar traffic flows with non-Gaussian probability distributions, in particular, long-tale distributions.
2. For the first time, an algorithm for determining overloads based on the information criterion was developed. As a criterion, we suggest using the approximated entropy of time series parameters. The dependence of the entropy of distributions on the probability of successful data transmission of one of the network nodes was calculated. The influence of the distribution entropy on the required resource for data exchange is shown.
3. The method of adaptive formation of network traffic flows with indirect feedback has been improved. The method differs from the previously proposed ones in that it has a fundamentally expanded vector of control actions, as a result of which the need for an additional feedback channel is eliminated. 4. For the first time, a method of optimizing the parameters and structure of the network traffic shaper was developed with the control of the length of the intervals of exceeding the levels of the flow parameters and the introduction of an additional module for predicting the required buffer size according to changes in the intensity of incoming packets.
In the first section, an analysis of the current state of the problem of design, implementation and application of network traffic shapers, prospects for the implementation of a single information space in any network controlled by a network traffic shaping system, etc., was carried out.
The main causes of overloading were analyzed, based on the results of the analysis, the goal and objectives were formulated, and the mechanisms of network management were investigated, such as the management of network resources according to the standards of the TMN (Telecommunication Management Network) management concept. Mathematical models of network traffic were studied; special attention is paid to self-similar traffic statistics with slowly decreasing time and frequency dependencies and probability distributions with "heavy tails".
The second section is devoted to methods of monitoring and analysis of network equipment, which are used to solve problems of managing network characteristics. A generalized model of managing the parameters of information flows in telecommunication networks has been built. Simplifying assumptions regarding the initial conditions of the network segment's functioning are formulated. An algorithm for determining network anomalies based on the entropy of time series has been developed. It is shown that the rate of growth of the required amount of memory in switching nodes increases with an increase in the Hurst parameter. The traffic shaper with variable speeds of arrival and processing of packets is considered. The methods of adaptive formation of network traffic flows and methods of setting up control structures of systems with indirect feedback, which control the parameters and structure of the shaper, are considered. A general transmission efficiency functionality with core and additional key network functions is developed.
The scheme of the M-th order adaptive traffic shaper using a modified prediction module based on the Smith predictor was developed. It is shown that with power-law smoothing and two-step forecasting, there is a decrease in the frequency drop interval by approximately 25% compared to exponential smoothing.
