Network traffic forecasting based on the canonical expansion of a random process

• Main Authors: Vitalii Savchenko, Oleksander Matsko, Oleh Vorobiov, Yaroslav Kizyak, Larysa Kriuchkova, Yurii Tikhonov, Andrei Kotenko
• Format: Article
• Language: English
• Published: PC Technology Center 2018-05-01
• Series: Eastern-European Journal of Enterprise Technologies
• Subjects: network traffic; forecasting control; random process; canonical decomposition of a random process
• Online Access: http://journals.uran.ua/eejet/article/view/131471

We studied the problem of forecasting network traffic in TCP/IP networks based on statistical observational data. We determined that existing protocols (SNMP, RMON) do not provide long-term forecasting, which is necessary for network upgrades. Regression methods (AR, ARMA, ARIMA, SARIMA), which are the basis of protocols, use only a sequence of values of forecasted series, which makes long-term forecasting impossible. We made a conclusion that there is no universal effective method for forecasting time sequences that describe traffic of a computer network. We developed the model of a forecast of network traffic taking into account features of accumulation of statistical data: presence of a priori trajectories, a posteriori character of forecasting, finiteness of variance. We applied the apparatus of the canonical expansion of a random process, taking into account heterogeneity of traffic. We developed a mathematical apparatus to solve the problem of extrapolation of implementation; we obtained expressions for the estimation of an extrapolation error, and expressions for the reconstruction of a posteriori random process based on modeling. We took into account accuracy of a priori measurements, which makes it possible to use this model in networks with a minimum of diagnostic data. It provides accurate determination of parameters of a random process at control points and the minimum standard approximation error in the intervals between these points. Application of the proposed method based on the canonical decomposition of random processes provides a solution to the problem of long-term forecasting of network traffic. A comparative analysis of forecasting methods indicates that the method of canonical decomposition of a random process comes close to intelligent forecasting methods.