Experience report on early software reliability prediction and estimation

The reliability growth of the network software for a multiship naval system was estimated using the Computer Aided Software Reliability Estimation (CASRE) program developed by A. Nikora (1994). Although CASRE provides useful results it isn't applicable until after system integration testing begins. To see if results could be obtained sooner, the CASRE estimates were compared with those from two other models that promised earlier results. The first was a development process model created by S. Keene. This model makes reliability predictions based on the Software Engineering Institute's Capability Maturity Model level of the developer and the estimated lines of code. The second model, developed by J. Gaffney and implemented in the Software Error Estimation Program (SWEEP) fits errors discovered to a Rayleigh curve. SWEEP was used to show our reliability growth from unit test forward. After all the results were obtained they were normalized to software errors remaining per thousand source lines of code and plotted on one graph. From this graph it was clear that software reliability could be predicted well before system integration.