Timing analysis by model checking

Abstract

The safety of modern avionics relies on high integrity software that can be verified to meet hard real-time requirements. The limits of verification technology therefore determine acceptable engineering practice. To simplify verification problems, safety-critical systems are commonly implemented under the severe constraints of a cyclic executive, which make design an expensive trial-and-error process highly intolerant of change. Important advances in analysis techniques, such as rate monotonic analysis (RMA), have provided a theoretical and practical basis for easing these onerous restrictions. But RMA and its kindred have two limitations: they apply only to verifying the requirement of schedulability (that tasks meet their deadlines) and they cannot be applied to many common programming paradigms. We address both these limitations by applying model checking, a technique with successful industrial applications in hardware design. Model checking algorithms analyze finite state machines, either by explicit state enumeration or by symbolic manipulation. Since quantitative timing properties involve a potentially unbounded state variable (a clock), our first problem is to construct a finite approximation that is conservative for the properties being analyzed-if the approximation satisfies the properties of interest, so does the infinite model. To reduce the potential for state space explosion we must further optimize this finite model. Experiments with some simple optimizations have yielded a hundred-fold efficiency improvement over published techniques.