Statistical abstraction boosts design and test efficiency of evolving critical systems

Abstract

Monte Carlo simulations may be used to efficiently estimate critical properties of complex evolving systems but are nevertheless computationally intensive. Hence, when only part of a system is new or modified it seems wasteful to re-simulate the parts that have not changed. It also seems unnecessary to perform many simulations of parts of a system whose behaviour does not vary significantly. To increase the efficiency of designing and testing complex evolving systems we present simulation techniques to allow such a system to be verified against behaviour-preserving statistical abstractions of its environment. We propose a frequency domain metric to judge the a priori performance of an abstraction and provide an a posteriori indicator to aid construction of abstractions optimised for critical properties.