accompanied by a sensitivity analysis comparing a collection of plausible priors in a given problem, what some think of as the “arbitrariness” of Bayesian analysis can be minimized (see, e.g., Samaniego and Reneau [1994], for results and discussion on this issue). Second, the use of methods for data combination by no means precludes the investigation of failure sources and the various steps that are associated with reliability growth management. Samaniego suggested that it is critical for theoretical and applied research to proceed simultaneously and interactively on both of these fronts.

In the floor discussion, Larry Crow asked whether anyone had compared failure modes in developmental versus operational test. Specifically, he wanted to know how much of the increased failure rate in operational test is due to new failure modes and how much is due to higher failure rates of known failure modes. Jim Streilein responded that his agency has an idea of what λ is for some systems. He added that efforts are being made to carry out failure modes and effects analyses, but the problem is that these estimates are never fully cognizant of the environment of use. Further, for most components, there is no physical model that can be used to provide reliability estimates for a given material or manufacturing process. Streilein is therefore not sanguine about combining information models until more information is available.

TWO MODEL-BASED APPROACHES TO SELECTING INPUTS FOR SOFTWARE TESTING

It is well known that software is essentially a ubiquitous component of today’s complex defense systems and that software deficiencies are a primary cause of problems in defense system development (see Mosemann, 1994). Software reliability, while sharing some aspects of hardware reliability, is clearly distinct in critical ways. For example, the smallest change to a software system can have a dramatic impact on the reliability of a software system. Further, there are no analogous notions to burn-in or fatigue for software components. Given the distinct nature of problems involving software and the broad aspects of the subject area, the hope during the planning stages of this workshop was that an entire subsequent workshop would be devoted to this issue, for which two presentations on software engineering at this workshop would serve as a preview. (This workshop was held July 19–20, 2001.)

This preview session outlined two approaches to the selection of in-



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement