Decomposition models for predicting software reliability
Noushin Ashrafi, Roger C. Baker · 1989
An examination of existing software reliability models reveals these models have deficiencies regarding either their assumptions or their practicality. Thus, there is a need for a new model which minimizes these deficiencies, permitting the software developer to predict software reliability with a higher degree of confidence. This study has developed three models: the Probabilistic Decomposition Model, the Deterministic Decomposition Model, and the Extended Bayesian Model. All Three models were developed to produce point estimates for software reliability, while the third model also produces a prediction interval for software reliability. The Probabilistic Decomposition Model makes use of the Markov Chain to represent the dynamic behavior of the transfer of control among the modules. The Deterministic Decomposition Model is only applicable when the user profile is known with certainty, in that the transfer of control between the modules is static. Both models describe software reliability as a function of its components: module reliability, interface reliability, and transfer of control probability. A numerical example was provided to demonstrate the application of both models. The Extended Bayesian Model utilizes Bayesian theory to combine a prior belief about the reliability of each module with additional data, obtained from module testing, to produce posterior distributions for each module. Monte Carlo simulation was employed to combine the posterior distributions for modules with other components affecting software reliability to produce a prior distribution for the software system. A posterior distribution for the software system reliability was developed by combining the prior distribution for the software system and sample data resulting from system testing. Sensitivity analysis was performed to examine the sensitivity of the reliability of the software system to the reliability of each module. Since the mathematical structure of the models is complex and its implementation is time consuming and tedious, computer programs were written for all three models.