A software-based fault injection tool (SOFIT)

Fault injection has long been recognized as a valuable approach for validat ing fault tolerance. This validation is based on fault injection that estab lishes a link between the behavior of the program and the fault occurrence process. In the past few years, many software-based fault injection tools h ave appeared in the literature. These tools have been developed with specif ic objectives that varied from (1) detecting design/implementation errors, (2) bringing a system to a specific state, (3) tracking error-propagation i n a system, and (4) estimating the dependability parameters for different f ault-tolerance techniques. Given the wide range of available architectures, operating systems, and the objectives of the designers, developing an all- purpose fault injection tool does not seem to be practical. New software-ba sed fault injection tools, developed primarily to meet the specific objecti ves of their designers, will continue to appear in the literature. One conc ern from the developer's point of view is how to develop a tool that could be portable across a variety of architectures. This paper presents SOFIT, a software object-oriented fault injection tool which has been developed for a Spare workstation running Solaris 2.x and ported to the nCUBE 2 parallel computer.