Statistical Measures of Software Reliability,
Abstract
Estimating program reliability presents many of the same problems as measuring software performance and cost: the central technical issue concerns the existence of an independent objective scale upon which may be based a qualitative judgement of the ability of a given program to function as intended in a specified environment over a specified time interval. Several scales have already been proposed. While these concepts may have independent interest, they fail to capture the most significant aspect of reliability estimation as it applies to software: most software is unreliable by these standards, but the degree of unreliability is not quantified. A useful program which has not been proved correct is unreliable, but so is, say, the null program (unless by some perversity of specification the null program satisfies the designer); an operationally meaningful scale of reliability should distinguish these extremes. In the sequel, we will sketch the outlines of the traditional theory that is most relevant to software reliability estimation, give a brief critical analysis of the use of the traditional theory in measuring reliability, and describe another use of the R(t) measure which we believe more closely fits the intuitive requirements of the scale we asked for above.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1980
- Accession Number
- ADA100662
Entities
People
- Frederick G. Sayward
- Richard A. Demillo
Organizations
- Georgia Tech