Software Defect Characterization by Growth Curve and Defect Type

This paper presents an empirical study on defect types and their impact on reliability growth, with a goal to enhance the design and development process. Defects are separated into subpopulations with varying reliability growth characteristics and analyzed for their defect type distribution. The defect type distribution is generated by analyzing a large sample of defects reports. The distribution is useful not only to relate cause to effect, but also to identify areas where the development process can be improved. We find that: . The population of defects is separable into sub-populations demonstrating different reliability growth characteristics, using failure symptoms. . Initialization defects are found to be strongly related to defect sub-populations with very inflected growth curves. This impacts the testing strategy. . Function and checking type defects contribute the largest cause for defects. However, improvements in specification can reuce the incidence of these defects, especially due to missing and incorrect (function or checking) types accounting for 40% of all defects.