Reliability prediction tools are design aids that assist engineers in selecting the appropriate system design. These tools aid engineers in comparing design alternatives and determining the impact of system parameters on the chosen design approach. This paper discusses the limitations imposed on reliability prediction tools by mathematics and the choices of their developers. A brief list of the limitations and capabilities of three current reliability prediction tools is presented in the paper. The paper also discusses four ways of overcoming reliability prediction tool limitations: selecting complementary tools, improving our understanding of computer system behavior, expanding reliability prediction tool capabilities, and employing a paradigm of reliability prediction tool use.With ever increasing frequency, computer systems are being developed for use in areas where their dependable operation is important. For example, computer systems are being used to control the transfer of billions of dollars in the banking industry, the operation of spacecraft in the aerospace industry, and the operation of X-ray and other therapeutic equipment in medicine, The failure of the computer system in each of these cases could result in the loss of millions of dollars, the loss of critical information needed for national defense, or the loss of life.The design of computer controlled systems is a difficult task, typically orders of magnitude more complex than the systems which were traditionally used. The evaluation of computer controlled systems involves guaranteeing that the mechanisms that ensure dependable system operation have been properly incorporated and that the system meets the required performance and reliability.Several methods and tools are required to evaluate the effectiveness of these mechanisms. For example, several reliability prediction tools have been developed that aid engineers in quantitatively assessing the effectiveness of a system's fault tolerance mechanisms for handling various fault types.The selection and implementation of a poorly designed system can impact both the system's users and the general public. The system users' or public's acceptance of a technology, or its application, is dependent upon how well it functions. If poorly designed systems are built and distributed, the acceptance of a technology could be adversely affected. At worst, a series of accidents would severely undermine public confidence in a technology and could result in litigation.This paper is organized as follows: in Section 2, the usefulness of reliability prediction tools to aid engineers in selecting the appropriate system design is discussed. In Section 3, limitations of reliability prediction tools as design aids are commented on. Also presented in this section is a brief list of the capabilities and limitations of three current reliability prediction tools: CARE 111, HARP, and CRAFTS (Ref. 4, 5, 6 ) . In Section 4, four methods of overcoming the limitations of reliability prediction tools are discussed. In the final sec...
