A Hybrid Approach for Predicting Aging-Related Failures of Software Systems

Software aging is a chronic process that is hidden under system monitoring until a system failure occurs. Aging related failures (ARFs) are the result of a variety of complex factors. Therefore, how to precisely predict the ARFs for a running software system is a challenge problem. Previous studies typically predict ARFs by means of predicting the time to resource exhaustion (TTE), which adopts resource data as aging indicators to predict when the resource data achieve the preset threshold. However, the practical effect of prior approaches are far from satisfactory due to lack of effective aging indicators and difficult to set accurate threshold. In this paper, we propose a hybrid approach, which combines model and measurements to construct a probabilistic aging indicator. The aging indicator is a multifactorial aging indicator that is more effective than traditional ones. Moreover, the hybrid approach is threshold-free in ARFs prediction. We evaluate the hybrid approach in Data caching system and Media streaming system, the results show that the hybrid approach can achieve high precision and recall for ARFs prediction. Compared to previous approaches, our approach increases the prediction precision and recall significantly.