Fault Injection for Failure Prediction Methods Validation

Failure prediction methods are becoming sine qua non conditions for effective availability enhancement in complex computer and communication systems. Therefore, there is a growing need for validation, benchmarking and assessment of such methods on real industrial data. Our thesis is that the effectiveness of such methods can be significantly enhanced when combined with fault injection. Then, not only failures can be predicted but also potential root causes can be identified based on symptoms, which can be observed at runtime on the system. We first briefly introduce failure prediction and fault injection methods and then present a methodology for improving and validating failure prediction methods using fault injection. We anticipate that with our approach we will be able to more efficiently predict forthcoming outages and also identify root causes which in turn will enable effective recovery or failure avoidance and as a result substantially enhance availability.