Constrained software distribution for automotive systems

Abstract

A variety of algorithms and technologies exist to cope with design space exploration for software distribution in terms of real-time, embedded, multiprocessor, and mixed-critical systems. The automotive domain not only combines those domains but even introduces further constraints and requirements due to several design decisions, standards, or evolved methodologies. In addition, solutions are predominantly proprietary, often lack in perspicuity, and sophisticated approaches towards the comprehensive concern of constraints are rather rare. This paper presents typical constraints along with distributing automotive applications across the processing units of vehicles, outlines three software distribution methodologies based on the constraint programming paradigm, and evaluates those in comparison to related design space exploration approaches. Benchmarks upon hypothetical and industrial models show that the constraint-based approaches outperform other forms in many cases regarding quality and effectiveness. Additionally, the presented approach benefits from a holistic consideration of constraints such as processing unit affinities, safety level aggregations, communication costs as well as processing unit utilization optimization among others whilst being applicable to heterogeneous, networked, hierarchical, embedded, multi and many core architectures.