Stochastic Contracts for Runtime Checking of Component-based Real-time Systems

摘要

This paper introduces a new technique for dynamic verification ofcomponent-based real-time systems based on statistical inference. Verifyingsuch systems requires checking two types of properties: functional andreal-time. For functional properties, a standard approach for ensuringcorrectness is Design by Contract: annotating programs with executable pre- andpostconditions. We extend contracts for specifying real-time properties. In theindustry, components are often bought from vendors and meant to be usedoff-the-shelf which makes it very difficult to determine their execution timesand express related properties. We present a solution to this problem by usingstatistical inference for estimating the properties. The contract frameworkallows application developers to express contracts like "the execution time ofcomponent $X$ lies within $\gamma$ standard deviations from the mean executiontime". Experiments based on industrial case studies show that this frameworkcan be smoothly integrated into existing control applications, therebyincreasing their reliability while having an acceptable execution time overhead(less than 10%).