Real-time checkers: built-in-test for mission-critical software

Testing is one of the most critical and difficult activities undertaken during software development and maintenance. Despite its importance, only limited progress has been made in developing a theoretical basis for testing software. Proving that mission-critical software will execute correctly for all possible inputs has been shown to be infeasible or prohibitively expensive. The end effect is that the contribution of residual software errors to system faults or failures in the mission environment continues to be a problem for which practical solutions have not been forthcoming. An approach called Program checkers shows considerable promise towards increasing the effectiveness of software testing without prohibitively increasing system cost. In particular checkers prove that a program operates correctly on a particular input and thus can serve as an oracle to automatically verify test or operational results. Initial research by Dr. Manuel Blum at UC Berkeley produced an approach to developing checkers for algorithms implemented with precise arithmetic, but this proved to have limited applicability to mission-critical applications. However, a recent breakthrough produced a checker for a numerical algorithm implemented with limited precision arithmetic. This paper reports on our experience in applying this new type of checker to a Fast Fourier Transform (FFT) implemented for the specialized signal processor of an airborne radar system.