| Combinatorial testing is one of the effective software testing methods. By generating a small amount of test cases defined as covering array, it can maximize the t-way coverage of software factors’ interactions with a reasonable cost and thus effectively detect the failures triggered by these interactions.As the test suite of combinatorial testing, covering array’s generation is one of the key issues in combinatorial testing. Many mathematical methods, greedy algorithms and evolutionary search methods have been applied in this field. Since the performance of evolutionary search methods is significantly impacted by their configurable parameters, we take genetic algorithm, one of the typical evolutionary search methods, as an example to discuss the different influences of its five configurable parameters (population size, evolution generation, crossover probability, mutation probability, variants of the algorithm) on the performance of 2-way covering array generation. Meanwhile we design three classes of experiments to systemically analysis the influences of each of the configurable parameters and the interactions among them in discrete value range.Our goal is to answer the following three questions:(1) Does there exist an improved configuration of genetic algorithm for a particular 2-way covering array generation?(2) Does there exist a common improved configuration for all 2-way covering arrays generation and(3) If there exists no common improved configuration for all 2-way covering arrays, is there any recommended parameter values of the genetic algorithm’s configuration that could lead to its relatively better performance for 2-way covering array generation?Beyond that, the experiment methods we proposed in this paper can be combinated and then adopted in the study of other evolutionary search methods for covering array generation, thus, providing a general framework for the performance improvement of the evolutionary search methods. |
PDF Full Text Request