By: Marco Vieira
“Nothing can assure the absence of errors”. Runtime failure prediction is one of the prerequisites to prevent system failures and thus increase availability. The problem is that a large amount of field failure data is needed for training failure prediction algorithms and for selecting the variables that show the best failure symptoms. In this talk we will explore the applicability of software fault injection techniques to improve failure prediction algorithms. In particular, we will present a concrete example on using injection of software faults for identifying variables for failure prediction in a complex system running Windows Xp.