Methodology of analysis and optimization of real-time embedded systems : application to automotive field
Ahmed Daghsen · HAL (Le Centre pour la Communication Scientifique Directe) · 2013
Today, the design and development of automotive software system becomes very complex. This complexity is due to the high number of functions, execution codes and diversity of communication bus embedded in the vehicle. Also, the heterogeneity of the architecture makes the design of such system more difficult and time consuming. The introduction of Model-Based Development (MBD) in the automotive field promised to improve the development process by allowing continuity between requirements definition, system design and the distributed system implementation. In the same direction, the apparition of AUTOSAR consortium standardized the design of such automotive embedded system by allowing the portability of software functions on the hardware architecture and their reuse. It defines a set of rules and interfaces to design, interconnect, deploy and configure a set of application software components (SWCs). However, designing an embedded system according to AUTOSAR standard necessitates the configuration of hundreds of parameters and requires several software allocation decisions. Each decision may influence the system performance and also the development cost. This architectural complexity leads to a large design decision space which is difficult to be explored without using an analytical method or a design tool. We introduce in this thesis a methodology that permits to assist and help the system designer to configure an AUTOSAR-compliant system. This is based on the Design Space Exploration (DSE) framework that permits to evaluate and analyze several design alternatives in order to identify the optimal solutions. The DSE task relies on a multi-objectives evolutionary algorithm. The DSE could be performed for two purposes : (1) the mapping of SWCs to ECUs and the mapping of runnables (code entities) to OS tasks, (2) the configuration of the software parameters like OS tasks priorities and types. The flexibility and scalability of the DSE framework allows applying it to other description and modeling languages such as SysML/MARTE.