Software-driven systems have become a fundamental part of our daily existence. Their growth in the past decades has, however, also led to our almost complete dependence on them. Depending on the application, the failure of software-driven systems may simply be frustrating, could lead to higher costs, or even be life-threatening. Take, for example, the flight controls used in modern aeronautics or the monitoring systems in place at a nuclear power plant.

New programming methods, changes in software production, and the widespread use of tools have made software products today much less prone to error than was the case with unplanned ad-hoc programming.

A significant problem remains, however: How to measure software quality, i.e. how to define control parameters and make these quantifiable. Since the early 1970's models have been available for estimating and predicting the reliability of computer software. The majority of these models use the times when errors occur during the test phase(s). However, as evidenced by the multitude of models in use, there has yet to be a generally accepted model which is adequate for every project. Under normal circumstances it is impossible to determine in advance which model predicts actual failure times during testing of specific software.

The distribution of errors detected in a software test obviously depends on software characteristics (and indirectly on the software production process) as well as on attributes of the test process. Nevertheless, the software industry is now only beginning to be able to identify significant factors and their relationships.

This project will analyze the data available from existing software processes and testing methods, and relate this data to software reliability models in order to increase their practical usability and accuracy. The PETS Project is also expected to provide knowledge about which factors promote or impede the quality of the final product.