Predicting How Badly "Good" Software Can Behave

This paper presents a fault injection methodology that predicts how software will behave when com ponents of the software fail hardware components external to the software fail human factor errors occur and bad input is provided to the software and the software is executing in unlikely operational modes Because of the enterprise critical nature of many of today s software systems it is vital that these system are robust enough to handle problems that originate externally as well as the expected problems that will arise from internal defects Also this paper presents four cases studies that highlight the bene t of this analysis for both safety critical systems and non safety critical systems