Fault exposure ratio estimation and applications

Abstract

One of the most important parameters that control reliability growth is the fault exposure ratio (FER) identified by J.D. Musa et al. (1991). It represents the average detectability of the faults in software. Other parameters that control reliability growth are software size and execution speed of the processor which are both easily evaluated. The fault exposure ratio thus presents a key challenge in our quest towards understanding the software testing process and characterizing it analytically. It has been suggested that the fault exposure ratio may depend on the program structure, however the structuredness as measured by decision density may average out and may not vary with program size. In addition FER should be independent of program size. The available data sets suggest that FER varies as testing progresses. This has been attributed partly to the non-randomness of testing. We relate defect density to FER and present a model that can be used to estimate FER. Implications of the model are discussed. This model has three applications. First, it offers the possibility of estimating parameters of reliability growth models even before testing begins. Secondly, it can assist in stabilizing projections during the early phases of testing when the failure intensity may have large short term swings. Finally, since it allows analytical characterization of the testing process, it can be used in expressions describing processes like software test coverage growth.