Comparing the fault detection effectiveness of n-way and random test suites

Software testing plays a critical role in the timely delivery of high-quality software systems. Despite the important role that testing plays, little is known about the fault detection effectiveness of many testing techniques. We investigate "n-way" test suites created using a common greedy algorithm for use in combinatorial testing. A controlled study is designed and executed to compare the fault detection effectiveness of n-way and random test suites. Combinatorial testing is conducted on target systems that have been injected with software faults. The results are that there is no significant difference in the fault detection effectiveness of n-way and random test suites for the applications studied. Analysis of the random test suite finds that they are very similar to n-way test suites from the perspective of the number of test data combinations covered. This result concurs with other hypothetical results that indicate little difference between n-way and random test suites. While we do not expect this result to apply in all combinatorial testing situations, we believe the result will lead to the design of better combinatorial test suites.