Applying reliability models more effectively (software) - IEEE Software

Abstract

M ore than 40 software-reliability models have been created since the first one appeared in 1972.' As new projects arose, new reliability models were created to suit them. Now software engineers have a plethora of reliability models, none of whch work optimally across projects. Consequently, a major difficulty in software measurement is analyzing the context in which measurement is to take place to determine beforehand which model is likely to be tmstworthy. Because software development and operation involve many intricate human activities and because software failure pattems are uncertain, such determinations are difficult if not impossible. Also, project data varies considerably and often does not coniply with a model's underlying assumptions. Thus, practitioners have no reliable way of knowing in advance which model is likely to produce the most trustworthy predictions. Instead of developing more detaded and potentidy more complicated -models, we chose to focus on using existing models more effectively. To this end, we have developed a set of hear combination models that combine the results of single, or component, models. As measured by statistical methods fordeterminingamodel'sappliabilitytoaset of Mure data, a combination model tends to have more accurate short-term and longterm predictions than a component model. After evaluating these models using both historical data sets and data from recent Jet Propulsion Laboratory projects, we have found that they are consistently satisfactory. To make it easier to apply reliability models and to form combination models, we are developing a tool to auto' 8

Cite this paper

@inproceedings{2009ApplyingRM, title={Applying reliability models more effectively (software) - IEEE Software}, author={MICHAEL R . and Allen P. Nikora}, year={2009} }