Tackling the Complexity of Timing-Relevant Deployment Decisions in Multicore-Based Embedded Automotive Software Systems

Multicore-based ECUs are increasingly used in embedded automotive software systems to allow more demanding automotive applications at moderate cost and energy consumption. Using a high number of parallel processors together with a high number of executed software components results in a practically unmanageable number of deployment alternatives to choose from. However correct deployment is one important step for reaching timing goals and acceptable latency, both also a must to reach safety goals of safety-relevant automotive applications. In this paper we focus at reducing the complexity of deployment decisions during the phases of allocation and scheduling. We tackle this complexity of deployment decisions by a mixed constructive and analytic approach. On the constructive side, we model a multicore-based embedded automotive software system as a tuple S = (L, R) with a given architecture L of the system's software component network and a given architecture R of the system's platform component network. On the analytic side, we derive models of the system's deployment as a tuple D = (M, Z), where the relation M: L -> R describes the allocation mapping from software components to platform components and the relation (Z:L -> N)(0) describes the scheduled start times of software components. The architectures L and R allow to describe hierarchically nested software-and platform-component networks. Hence, we are able to represent coarse-grain structures like domain functions and domain networks as well as fine-grain structures like individual driver-calls, algorithms, ECUs and CPU-cores as well. With all this architectural information contained in precise models, deployment alternatives (allocations and schedules) can be automatically derived and analyzed for conformance with timing requirements, given by a set of timing constraints C. To demonstrate the usability of the approach, we implemented an architecture description language in the academic open source tool CADMOS together with a prototypical allocation and scheduling support. We are evaluating the practicability of the approach within a case study in the German national joint project ARAMiS using a close-to-production multicorebased system in cooperation with Technische Universitat Munchen, AUDI AG and BMW AG.