Software reliability prediction for large and complex telecommunication systems

The problem of predicting the number of remaining faults in a software system are studied. Seven software projects are analyzed using a number of software structure metrices and reliability growth models. The following conclusions are drawn: there is no single model that can always be used, irrespective of the project conditions; software structure metrics (mainly size) do correlate with the number of faults; the assumptions of reliability growth models do not apply when the testing is structured and well organized; and sufficient data has to be collected from different projects to create a basis for predictions