Refining a randomized post-optimization method for covering arrays

Combinatorial testing is an efficient technique that reveals the faults caused by parameters interaction in software systems. It uses covering arrays as test suites to avoid the combinatorial explosion of parameter values. However, the construction of covering arrays with minimum size is still a challenging problem. This is because most of the existing methods to construct covering arrays result in extensive repetition of coverage. To address this challenging problem, some researchers proposed a randomized post-optimization method to reduce the number of rows in a covering array that has been already constructed. It repeatedly adjusts the array without loss of coverage. However, it seems that this method is not efficient as desired. In order to improve its efficiency, we propose a refined algorithm to replace the randomized algorithm adopted in the method. This avoids the local optima problem that arises in randomized algorithm. Experimental results are presented to confirm the benefits of the refined algorithm. In the cases of CA(N, 6, k, 2) (38 ≤ k ≤ 50) generated by IPO, this refined algorithm not only reduces more number of rows than the randomized algorithm, but also produces covering arrays that are smaller than previously best known ones. Furthermore, we provide a reasonable estimator to predict the extent to which covering arrays can be improved by this refined algorithm.