SPLINE-EXTRAPOLATION METHOD FOR RESTORING SELF-SIMILAR TRAFFIC

Abstract

The problem of predicting self-similar traffic, which has a considerable number of bursts and ripples and the property of long-term dependence, using the method of spline-extrapolation using cubic B-splines was considered in this paper. The application of the spline-extrapolation method allowed to predict self-similar traffic outside the analyzed time segment, which deals with packet data transmission on the mobile network. A method for estimating the traffic predicting error using the method of spline-extrapolation based on cubic B-splines is proposed. According to the results of the study, it is found that cubic B-splines are characterized by high design and ease of practical implementation. Their use makes it possible to greatly simplify computational processes using formalized mathematical constructs. On the example of simulated self-similar traffic received in the Simulink package of the Matlab environment, a spline-extrapolation was performed using cubic B-splines. This made it possible to obtain a "route" of traffic beyond the considered segment, which corresponds to the results of modeling with a certain prediction accuracy. The use of the proposed method of spline-extrapolation based on cubic B-splines has several advantages over other known methods, namely, ease of practical implementation, high accuracy of forecasting, the ability to accurately extrapolate peak "bursts" of traffic, which is especially important when solving problems in real time. The practical significance of the results of the study is determined by the fact that the obtained values of traffic intensity with a known load of network nodes, will allow the operator at the stage of design and subsequent operation of the mobile network to predict the required volume of buffer devices network hardware, thereby avoiding network congestion and accidence of regulatory values of traffic service quality characteristics. The prospect of further research is suggested by using wavelet extrapolation to improve the accuracy of estimates predicting.