Understanding Ineffectiveness of the Application-Level Fault Injection

Extreme-scale applications are at a significant risk of being hit by soft errors on supercomputers, as the scale of these systems and the component density continues to increase. In order to better understand soft error vulnerabilities in those applications, the application-level fault injection is widely employed to evaluate applications. This poster reveals that the application-level fault injection has some inherent uncertainties due to the random nature of fault injection. First, the fault injection result has a strong correlation with the number of fault injection tests. What is a good number of fault injection tests is uncertain. Second, given a specific application, the fault injection result can vary as the input problem of the application varies. How to interpret the fault injection result is uncertain. Those uncertainties can make fault injection ineffective for accurately modeling application vulnerability.