On The Development Of Fault Injection Profiles

Abstract

The impact of hardware failures on software has attracted substantial attention in the study of dependable systems. Fault injection techniques have emerged as a major means to evaluate software behavior in the presence of hardware failures. However, due to the lack of knowledge of the fault distribution information, the fault location and time are randomly selected. One major drawback of this approach is that the injected faults do not represent the system's operational situation, thus software reliability cannot be credibly assessed. This paper aims at extending the use of fault injection to the reliability prediction of hardware faults. To do so, we have developed a set of analytical and simulation based methods capable of statistically reproducing the underlying physics and phenomena leading to hardware failures in a given system operational context. Such distributions are referred to as fault injection profiles, and are the basis to extend the fault injection technique with fault models that represent the actual conditions under which hardware faults occur