S. Mankefors-Christiernin and A. Boklund (Sweden)
Software testing, testing efficiency, random testing, op erational profiles, parallelization, integral formulation.
Using the equivalent integral formulation of random test ing, approximation of operational profiles and MPI-based parallelization, we present an integrated over view and evaluation of the efficiency of random testing with these improvements and compare them to traditional random testing. For the parallelization it is found that the effi ciency scales positively with the size of test cases, i.e. displays an almost linear speedup with respect to the number of computational nodes for larger software, while the profile approximation allows arbitrary many profiles to be evaluated at the constant cost of a single test suite. Combing parallel implementations and constant cost evaluations, a very extensive efficiency increase is found. In certain scenarios the efficiency per time unit goes up by a factor of 100 or more, making random testing into a practically viable testing technique. In the view of the straightforward nature of random testing these findings makes random testing a definitively interesting candidate for automatization in combination with other black box techniques.
Important Links:
Go Back
