Analysis and forecasting of computer network characteristics

Abstract

Today most applications and devices interact with the data networks, so it is important to develop and improve technologies and methods to better understand, control, manage or predict the behavior and state of computer networks and their characteristics. Therefore, the tasks related to the development of models and methods for evaluating and forecasting computer network traffic parameters are important for computer network management. The paper considers the traffic of computer networks in terms of the time series. Trend models of the time series, trend detection criteria and assessment methods are reviewed and analyzed. The selected method for evaluating the traffic trend is based on the Mann-Kendall test, and the consensus method has been used to interpret the results. The task of forecasting computer network traffic taking into account trend indicators also has been considered. This problem has been successfully solved by using a forecasting model based on a moving average, and improved by using gradient boosting. A separate task was to collect and pre-process set of input data describing the operation of a computer network, to formalize it and perform subsequent quantitative and qualitative analysis. A unique data set has been created by parsing logs (system files) from monitoring the state of computer network traffic. It is this set of data that has been used to create a trend detection model and further forecast the characteristics of the computer network. The obtained results shows that the developed models and methods can be used on practice solving problems of monitoring and managing computer networks.