Scheduling in critical real-time systems: a manifesto

Abstract

We investigate the nature of the scheduling problem as it arises in critical real-time systems, as well as the nature of its solutions. We establish that critical real-time systems must be designed as fault-tolerant distributed real-time systems. We argue that any provably correct solution to the scheduling problem must rest on on-line algorithms. Limitations of «solutions» currently used or proposed are examined. A design methodology is introduced, which serves the purpose of enforcing the timeliness proof obligation, such proofs being established under incomplete knowledge. Techniques that facilitate such proofs are pointed out and examples of solved problems are given.