Verifying Timing Constrains by means of Evolutionary Testing in a Nondeterministic Environment

Genetic algorithms can be used to automatically find test cases to test the temporal requirements of real-time systems. There is a current trend to use “off the shelf” hardware and software to implement soft real-time systems. These off the shelf products are designed to optimize the average performance which causes nondeterministic behavior. Genetic algorithms normally used to find test cases for testing temporal requirements do not work in a nondeterministic environment. This problem is identified and explained in this paper. Several solutions are proposed and tested. It is concluded that recalculating the fitness, for each individual, for each new generation is the best method when searching for a test case which will produce the WCET for a piece of code under test. The effect of a nondeterministic environment in the automatic search for temporal test cases is only studied for one algorithm (insertion sort) in one specific environment (Pentium 4 with Windows XP). More work is needed to verify the proposed solution for other algorithms and other environments.