Simple One-to-One Architecture for Parallel Execution of Embedded Control Systems

Abstract

We propose in this paper a simple architecture for efficient execution of embedded control systems using a model-based design to which automatic parallelization is also applicable. This architecture makes use of the advantages of embedded control systems in the sense of parallel execution, and reduces their disadvantages by incurring minimal overhead from task scheduling and inter-task communication. With a model predictive control application, our architecture achieves communication latency 20 times faster than in current realtime OS communication methods for many cores. Moreover, our architecture achieves a speed-up for 64 cores that is 40 times that in single core execution, while its performance scalability is saturated up to 32 cores in real-time OS communication.