Application-Specific Architecture Selection for Embedded Systems via Schedulability Analysis

Abstract

Architecting real-time embedded systems is of the top significance during the design phase, especially in complex applications. Due to limited time and resource, to guarantee scheduling eminence without violating application-specific constraints is a challenging problem in architecture level. In this paper, we firstly present an enhanced transformation from AADL models to Cheddar input for schedulability analysis. With subprogram and delayed connection, this transformation is feasible for complex system designs. Based on schedulability analysis, we further propose a novel architecture selection engine, which evaluates scheduling performance through selection standards and application-specific constraints via satisfaction functions. With the proposed selection engine, information from both schedulability and real-time constraints are captured to pick up an optimal architecture. We apply the proposed approach on the architecture selection of an industrial control system in railway applications. Four candidate AADL architectures are transformed and analyzed for schedulability. Then in the selection engine, candidates are ranked within two application constraints. Compared to the selection of general criteria and traditional AHP, our engine excels at better schedulability and satisfaction on real-time application-specific constraints. Moreover, with adjustment on constraints, our engine shows delicate sensitivity by generating a modified selection. We believe the proposed approach can facilitate architecture design of real-time embedded systems.