Advanced Failure Prediction in Complex Software Systems

摘要: The availability of software systems can be increased by preventive measures which are triggered failure prediction mechanisms. In this paper we present and evaluate two non-parametric techniques model predict the occurrence failures as a function discrete continuous measurements system variables. We employ modelling approaches: an extended Markov chain approximation technique utilising universal basis functions (UBF). presented methods data driven rather than analytical handle large amounts variables data. Both have been applied to real commercial telecommunication platform. includes event-based log files time continuously measured states. Results in terms precision, recall, F-Measure cumulative cost. compare our results standard such linear ARMA models. Our findings suggest significantly improved forecasting performance compared alternative approaches. By using may order magnitude.