Mathematical Models and Methods for Monitoring and Predicting the State of Globally Distributed Computing Systems

Abstract

Monitoring events and predicting the behavior of a dynamic information system are becoming increasingly important due to the globalization of cloud services and a sharp increase in the volume of processed data. Well-known monitoring systems are used for the timely detection and prompt correction of the anomaly, which require new, more effective and proactive forecasting tools. At the CMG-2013 conference, a method for predicting memory leaks in Java applications was presented, which allows IT teams to automatically release resources by safely restarting services when a certain critical threshold value is reached. Article’s solution implements a simple linear mathematical model for describing the historical trend function. However, in practice, the degradation of memory and other computational resources may not occur gradually, but very quickly, depending on the workload, and therefore, solving the forecasting problem using linear methods is not effective enough.