A Model for T-Way Fault Profile Evolution during Testing
D. Richard Kuhn, Raghu N. Kacker, Yu Lei · 2017
Empirical studies have shown that most software interaction faults involve one or two variables interacting, with progressively fewer triggered by three to six variables interacting. This paper introduces a model for the origin of this distribution. We start with two empirically reasonable assumptions regarding the distribution of branch conditions in code and the proportion of t-way combinations seen in testing, and show that the model closely reproduces empirical data on t-way fault distributions. Although the model was developed to explain the distribution of faults by t-way interaction strength, it is shown to reproduce the basic exponential reliability model as a special case at each level of interaction, t. The paper evaluates model predictions against empirical data, and discusses implications for detection and removal of interaction faults.