A field data study on the use of software metrics to define representative fault distribution

Authors

Abstract

This paper evaluates the use of software complexity metrics to define representative fault distributions for software fault injection experiments. A field data study on more than 350 bug reports available from open software initiatives was used to compare the fault distributions generated by our approach with the real fault distributions observed in the field. Results show that the way we distribute software faults for fault injection is consistent with field observations when the size and complexity of the software modules are not very high. For very large modules the fault density observed are lower than the estimated by our approach. Possible explanations and improvements are discussed.

Conference

Workshop on Empirical Evaluation of Dependability and Security (WEEDS), June 2006