G EN EE

Abstract

To guarantee the safe completion of a specific task in an energy-constrained (i.e., battery-operated, energy-harvesting) cyber-physical system (CPS), information on the task's worst-case energy consumption (WCEC) is necessary. To determine upper bounds on the WCEC, analysis tools conducting static program-code analysis make sound but conservative assumptions on the program's dynamic behavior. When using existing source-code benchmarks, knowledge of their possible program paths and thus their dynamic behaviors are not available. This lack of knowledge leads to missing baselines, which prevents comprehensive evaluations of the accuracy of WCEC analyzers, that is, the difference between the actual WCEC and analyzer's reported upper bound is unknown.In this paper, we present GENEE, a benchmark generator that enables in-depth evaluations of WCEC analysis tools for energy-constrained CPSs. GENEE combines small program blocks in a way such that all necessary characteristics of the benchmark are available after the generation process together with the benchmark. Since peripherals (e.g., transceivers) are of major importance when analyzing the energy consumption of CPSs, GENEE generates benchmarks that (de-)activate these components while tracking all possible dynamic behavior. Evaluations with our open-source prototype of GENEE on an ARM Cortex-M4 platform show that the generator produces programs with device interactions and known characteristics, most importantly, the program's actual WCEC.