On Basis Variables for Efficient Error Detection
Jake Fairbrother, Matthew Leeke · 2017
The development of dependable software invariably entails the design and location of error detection mechanisms. This software artefact type captures predicates over program variables in order to facilitate error detection. To ease the design of detectors, it is important to have (i) knowledge of the set of variables to be included in a predicate and (ii) an understanding of the structure of the predicate. In this paper, we address these problems by relating a previously defined software metric to the variables that feature in efficient error detection predicates. Specifically, based on fault injection analysis of three software systems, we show that error detection predicates based on the 25% most important variables in a software module provide a similar level efficiency to those predicates that are based on all variables and variables with high importance value appear at lower depths in the generated decision tree, thus implying that these variables provide the most information with regard to system failure and, hence, should be protected to provide proper software function. The implication of these results is that, in order to develop efficient error detection predicates, it is sufficient to only have knowledge of a basis set of important variables, simplifying the design of efficient detectors.