Dependent input sampling strategies : using metaheuristics for generating parameterised random sampling regimes

Understanding extreme execution times is of great importance in gaining assurance in real-time embedded systems. The standard benchmark for dynamic testing'uniform randomised testing'is inadequate for reaching extreme execution times in these systems. Metaheuristics have been shown to be an effective means of directly searching for inputs with such behaviours but the increasing complexity of modern systems is now posing challenges to the effectiveness of this approach. The research reported in this paper investigates the use of metaheuristic search to discover biased random sampling regimes. Rather than search for test inputs, we search for distributions of test inputs that are then sampled. The search proceeds to discover and exploit relationships between test input variables, leading to sampling regimes where the distribution of a sampled parameter depends on the values of previously sampled input parameters. Our results show that test vectors indirectly generated from our dependent approach produce significantly more extreme (longer) execution times than those generated by direct metaheuristic searches.