Quantifying the reliability of computer software

Although models for quantifying software reliability are a controversial subject , they are the only ones that provide an objective assessment of the level of reliability achieved at the end of development. As a result, this point is never explicitly addressed, or is approached subjectively, or is replaced by requirements on the realization process. In all cases, the substitution techniques used do not provide a precise idea of the risk involved in putting a software product into operation, which is, in essence, the aim of reliability quantification.

This article describes the theoretical foundations of software reliability quantification and explains how it can be used. It endeavors to dispel the main misgivings about software reliability quantification, explaining in particular the process by which failures occur and how mathematicians have proposed to model it. He also explains how OEMs could implement software reliability requirements that would clearly contribute to better operational quality of the IT components of the complex systems they acquire.

The aim of the article is to show that reliability modeling is a valid statistical technique applicable to software of all kinds: embedded software, real-time software, information system software in all fields. The article also takes stock of current industrial practices, and suggests possible future developments.