A penalty-based Tabu search for constrained covering arrays.

Combinatorial Interaction Testing is a black-box testing technique particularly used for highly configurable software systems, which involve a number of factors (and values) that can be combined, according to some constraints. In this context, constrained covering array (CCA) is a central combinatorial problem tasked with building a test suite of minimum size and maximum coverage of the factors' interactions. In this paper, we propose CATS (Covering Array by Tabu Search), a new penalty-based tabu search algorithm for the CCA problem. Our local search approach differs from the ones previously proposed primarily by its use of a search space that allows solutions that violate inter-factor constraints. Other prominent features of CATS are the definition of strategic moves used to restrict the neighborhood, and a technique to vary the tabu tenure throughout the search. We performed tests with CATS on 2-way constrained problems using 35 widely used benchmarks. Results suggest that CATS consistently outperforms previous approaches, both on the size of the test suites and the needed computation times.