The Virtues of Assessing Software Reliability Early

A t the start of my academic career, a manager of a company that provides independent software verification and validation services to NASA approached me with a deceivingly simple question: Is it possible to predict software reliability before the system is completely implemented and integrated? The manager's question likely came from his knowledge of independent verification and validation's specific practices, which include rigorous analysis and quality-improvement activities applied early and throughout the development life cycle. Given these practices , the ability to find out early on whether a project will reach its target reliability is a logical aspiration. Software reliability is one of the few software quality attributes with a sound mathematical definition: the probability of a software failure's occurrence within a given period and under specific use conditions. By this definition , reliability is a strictly operational quality attribute. Traditional software-reliability prediction methods such as reliability growth models base estimates on observing failures (and fixing faults) in validation testing, during which operational patterns represent the product's actual field use. However, most traditional methods ignore all the quality indicators collected before system integration. A reliability prediction method that integrates quality information from such sources as architectural system descriptions, use scenarios , system deployment diagrams, and module testing lets managers identify problem areas early and make any necessary organizational adjustments. Few software projects these days start development from scratch. We create systems by evolving existing ones, such as extending product lines, adding COTS modules, or just including library functions. Some components come with significant quality-related information; others come with almost nothing. An architectural drawing board can provide quality engineers with valuable information—if only they could piece together the amazing puzzle of seemingly unrelated quality indicators. Yet, when it comes to software reliability modeling techniques, most engineers don't consider the specifics of component-based and reuse-oriented software-development paradigms. Several techniques analyze component-based applications. These methods all aim to calculate application reliability using the reliability of individual components and their interconnections. We can broadly classify component-based estimation techniques as either state-or path-based models. 1 State-based models use control graphs to depict application architecture. Software analysis tools extract these graphs from the application code. Analysts can transform a program control graph into a Markov model—a state-based rep