Theory of System Reliability Based on Components1

Most software-component research has been directed at functional specification of software components. The other, equally important, side of the coin is component quality. We present a foundational theory of reliability based on components. The theory describes in principle how component developers can make measurements that are later used by system designers to calculate — without implementation and test — system reliability. The theory is a ”microscopic” one that describes in detail how component properties are reflected in systems designed using those components. 1 PITFALLS OF SOFTWARE COMPONENTS Software components are the most promising idea extant for the efficient design of quality software systems. Most of the research in components is devoted to specification, design, reuse, and cataloging of the components themselves. The complementary issue – component quality – is also important, but has received less attention. There are no accepted standards for the quality of software components, largely because there is no theoretical foundation on which to base standards. Developers of safety-critical software, and the regulatory agencies responsible for the systems they design, use mostly subjective assessments of software quality. It would be of great value to replace these with hard data. In electrical and mechanical engineering, components are described in a handbook, where each has a “data sheet” entry. Its data sheet describes what a component does, and equally important, it gives constraints that allow the system designer to decide if the component is “good enough” for the application. For mechanical components, these constraints concern, for example, the life expectancy of the component. Software is embedded in systems with mechanical and electrical components, systems designed using component techniques from these other branches of engineering. The system designer using an embedded software system would like the software components to have data sheets. Without the solid information of a data sheet, software components may be no bargain. To buy off-the-shelf software of unknown quality is only to trade the difficult task of assessing your own work, for the more difficult task of assessing someone else’s. Software reliability theory [10] is a candidate for describing “quality” on a component’s data sheet, but it cannot be applied without addressing a central problem. Whereas reliability of mechanical components depends on their physical environment, and can be given without regard for the expected usage so long as that usage remains within a A short abstract describing some of this work appeared in ISSRE ‘99, Boca Raton, FL, Nov 1999; it was also used in a position paper for Workshop-12 in OOPSLA-99, Denver, CO, Nov 1999.