Predicting Software Reliability

• Authors: Alan Wood
• Venue: Computer
• Year: 1996

Software vendors need to know if their products are reliable before they are delivered to customers. Software reliability growth models help provide that information. Unfortunately, very little real data from commercial applications have been published about the utility of these models, possibly because of the proprietary nature of the data. Using transformed data to protect proprietary information, this article reports Tandem's experience with software reliability growth models. The data involves products for four major software releases. The author and his colleagues at Tandem collected defect occurrence times during system test and statistically correlated the data with known mathematical functions, using eight software reliability growth models. The authors found that, although they are still in the experimental stage, software reliability growth models can be used to provide reasonable predictions of the number of defects remaining, which is an indication of whether software is ready to release to customers. The results show that predictions from simple models of defect occurrence times correlate reasonably well with field data. However, the many choices for data representation and model type must be evaluated across multiple software releases to determine the appropriate models and obtain confidence in the results.