Using Holt - Winters models for forecasting the performance of server systems

The application of the Holt - Winters model for predicting the performance of server systems is analyzed. It is noted that for this purpose it is necessary to collect statistical information about the parameters of objects representing time series. The specifics of the time series of server systems to select the appropriate forecasting model are described. Autoregression models, neural networks, and exponential smoothing are compared for application to the mathematical problem posed. It has been argued that the Holt - Winters model has the advantage of analyzing the series of a group of servers containing tens / hundreds of parameters. Experimental studies are also carried out to assess the accuracy of the Holt - Winters model with respect to time series describing the dynamics of changes in server parameters. It is concluded that the triple exponential smoothing model shows good results and can be used to solve practical problems.