Anatomy of safety-critical computing problems

This paper analyzes the obstacles faced by current safety-critical computing applications. The major problem lies in the difficulty to provide complete and convincing safety evidence to prove that the software is safe. We explain this problem from a fundamental perspective by analyzing the essence of safety analysis against that of software developed by current practice. Our basic belief is that in order to perform a successful safety analysis, the state space structure of the analyzed system must have some properties as prerequisites. We propose the concept of safety analyzability, and derive its necessary and sufficient conditions; namely, definability, finiteness, commensurability, and tractability. We then examine software state space structures against these conditions, and affirm that the safety analyzability of safety-critical software developed by current practice is severely restricted by its state space structure and by the problem of exponential growth cost. Thus, except for small and simple systems, the safety evidence may not be complete and convincing. Our concepts and arguments successfully explain the current problematic situation faced by the safety-critical computing domain. The implications are also discussed.