An Empirical Analysis of Providing Assurance for Self-Adaptive Systems at Different Levels of Abstraction in the Face of Uncertainty

Abstract

Self-adaptive systems (SAS) must frequently continue to deliver acceptable behavior at run time even in the face of uncertainty. Particularly, SAS applications can self-reconfigure in response to changing or unexpected environmental conditions and must therefore ensure that the system performs as expected. Assurance can be addressed at both design time and run time, where environmental uncertainty poses research challenges for both settings. This paper presents empirical results from a case study in which search-based software engineering techniques have been systematically applied at different levels of abstraction, including requirements analysis, code implementation, and run-time validation, to a remote data mirroring application that must efficiently diffuse data while experiencing adverse operating conditions. Experimental results suggest that our techniques perform better in terms of providing assurance than alternative software engineering techniques at each level of abstraction.