Predicting web application crashes using machine learning

Unplanned system outages have a negative impact on company rev- enues and image. While the last decades have seen a lot of efforts from industry and academia to avoid them, they still happen and their impact is increasing. Ac- cording to many studies, one of the most important causes of these outages is resource exhaustion for different reasons: overload, inadequate system resource planning, or transient software failures which consume resources until crash. Sev- eral previous work have proposed the use of machine learning algorithms for modeling and predicting resource consumption, and the effectiveness of these ap- proaches have been demonstrated in failureless, stationary circumstances. In this paper, we present a framework based on machine learning techniques to predict the time to crash when the system suffers transient software errors such as mem- ory leaks which consume resources randomly and gradually. The experiments illustrate that our approach is effective at predicting crashes (at least of a particu- lar type, those due to memory leak) and with a lot of potential impact. Moreover, we present a discussion about the monitoring systems granularity level.