Error models for the representative injection of software defects

Abstract

This paper considers the representativeness of injected error models for ascertaining software defects. Businessand safety-critical systems are more and more relying on software. Therefore, while in the past these systems were mainly threatened by hardware faults, they are today increasingly exposed to software faults, as demonstrated by recent severe software-related accidents [WDS+10]. It is a matter of fact that, despite careful engineering and rigorous quality assurance, critical systems are deployed with residual (unknown) software defects. This problem is exacerbated by the massive reuse of legacy and off-the-shelf software components [Wey98, Voa98]: When a component is reused in a new context, the system may use parts of the component that were previously seldom used and only lightly tested, or may interact with the component in unforeseen ways, thus exposing residual software faults in the component that had not been discovered before. It thus becomes important to adopt software fault tolerance strategies, in order to prevent such residual defects in less critical parts from affecting more critical parts of a system. Software fault injection (SFI) is an experimental approach to assess the dependability of software-intensive systems in the presence of faulty software components, and to guide the development of software fault tolerance mechanisms and algorithms. SFI deliberately