In Professor Bernstein's paper on fault tolerance he discusses a classical engineering technique that involves stressing a system until it breaks and then certifying it for considerably less than the breaking point. I was wondering to what extent this technique is used in software testing and certification. How many of you have done this in any system that you have produced either academically or in industry? I estimate that less than half of the systems I have been involved with have: (1) been stressed and (2) certified for much less than the stress point. This is despite the fact that in most instances of systems that we stress tested, it avoided a major fiasco.
So the open question is how many of you stress test each system and, if so, do you certify the system, or make sure it operates within, parameters much lower than the stress point?