All the features implemented for a stabile optimization, like rescheduling of failed tasks are useless if the simulators used for model function are numerically instable. In interactive use of these tools these problems can be solved by tuning some parameters or the input data, in an optimization loop this is not practicable. On the other hand the several thousand executions of a simulator, for example MINIMOS-NT, have discovered some faults and so improved the quality of the program.