Software reliability and safety

Citation

Rudall, B.H. (1999), "Software reliability and safety", Kybernetes, Vol. 28 No. 5. https://doi.org/10.1108/k.1999.06728eaa.004

Software reliability and safety

Keywords Automation, Cybernetics, Research, Technological developments

Abstract Reports and surveys are given of selected current research and development in systems and cybernetics. They include: Language interface, Automated automobile, Innovative space technology, Software reliability and safety, Automatic analysis of handwritten documents, High-tech musical instruments, Biological motors, Interplay between smell and the mind, Cybernetics and automation.

Software reliability and safety

Centre for Software Reliability

City University UK decided in 1983 to set up a Centre for Software Reliability (CSR). It recognised the need for dealing with software in a scientific fashion at a time when developing computer programs was an art reserved for the small number of computer "buffs" who more often than not kept mathematics and science well away from their products. Since that time the Centre has developed novel and sophisticated statistical methods for analysing how a system's software reliability grows as faults are identified and removed.

Measuring and prediction techniques

The importance of developing, measuring and predicting techniques for software reliability and safety are now recognised. Techniques pioneered at CSR that accurately measure and predict software dependability are being broadened and made more accessible to users through a number of the current UK's Engineering and Physical Sciences Research Council (EPSRC) funded projects. CSR's Director, Professor Bev Littlewood says that:

Measuring a program's current reliability and trying to predict its future reliability growth turn out to be surprisingly difficult. There are several sources of uncertainty. For instance we do not know which input sequences will induce failure, nor how fixes will affect subsequent reliability.

A report on the work of CSR published in Research File, Impact Update, EPSRC No. 22, March 1999, describes some of the themes and techniques developed by its researchers. One theme of CSR research has focused on enabling users not only to obtain more accurate measures of reliability, but also to give them confidence in the accuracy of these metrics.

Another major novelty is CSR's "recalibration" technique which learns from past prediction errors on a data set to improve future predictions for that set.

The aim has been to give easy access to these techniques, even to users without statistical expertise. This is being developed in the EPSRC-funded project called Pre-Exploitative Tools for Evaluating Reliability of Software (PETERS). This is led by Professor Littlewood who also is investigating how the techniques can be applied with the relatively "coarse" incomplete data available in everyday applications, such as with pre-release beta software. Professor Littlewood observes: "These techniques have much value. However, they might not be appropriate for safety-critical applications requiring very high levels of reliability because they rely solely on software failure data".

Other research projects

In the United Kingdom this problem is being addressed not only by CSR but also by two other EPSRC funded projects: IMPRESS and DISCS. These projects have the following study aims:

• •

  IMPRESS - aims to combine disparate evidence about dependability using Bayesian Belief Nets (BBNs) in order to support stronger claims about reliability predictions. BBNs are graphical networks that represent probabilistic relationships among variables. The main investigators for this project are Professor Norman Fenton and Dr Martin Nell.

• •

  DISCS - a study of design diversity as both a means to achieve high levels of dependability and as a way of making claims about a system's fitness for purpose. This project is in collaboration with Newcastle University UK and Ing. Lorenzo Strigini is the principal CSR investigator.

Other studies are also being carried out at CSR. One project is to study the efficiency of different software testing approaches. A report on this work was published at the 1997 International Conference on Software Engineering, Boston, USA. The report won the "best paper" award at the meeting. Further details of this work are on: www.csr.city.ac.uk which is the Web link for CSR at the City University, UK.