|
H. Hecht, "A Proposal for Standardized Software Dependability Data", ISSES
95, Montreal, Canada, August, 1995
Software that does not fail is essential to many current ìapplications of digital
technology, and this need will increase as more functions critical to the individual
or the community are being automated. Examples are intelligent highways, use of
robots in direct contact with the public, medical monitoring, and drug administration.
Based on current specifications for critical software, the maximum tolerable failure
rates for the most severe modes range from 10-7/hr for complex multi-computer installations
to 10-10/hr for simpler programs running on free-standing computers [FAA91, FAA88].
To demonstrate attainment of such low failure rates requires approaches that go
much beyond the established methods of software reliability estimation, e. .g, [MUSA87].
Indeed, a large fraction of the development cost of high integrity systems must
be devoted to the assessment of their dependability through software quality assurance,
inspection and test. Yet most serious students of the field agree that none of the
assessment methods and tools currently employed provides an estimate that has a
high correlation with the achieved dependability for the required high reliability
levels. Because we do not know what and how much the individual steps of the assessment
contribute to the validity of the estimate, there is a distinct possibility that
some of the effort was wasted. Similar uncertainty exists regarding the value of
development methodologies and tools to the attainment of extremely high probability
of success. The total development and assessment cost for high dependability systems
is so large that even a small reduction of the effort by replacement of ineffective
with more effective procedures willresult in large economic benefits.
Click to Download
|
|
