Edge AI in the computing continuum: Consistency and Availability at Early Design Stages

This paper explores the integration of Edge Intelligence (EI) with the coordination language LINGUA FRANCA (LF), leveraging the Consistency-Availability-Latency (CAL) theorem as the theoretical foundation to optimize Cyber-Physical Systems (CPS) design and deployment. We propose a distributed EIbased approach for CPS to develop an Emergency Vehicle Detection (EVD) system that dynamically adjusts traffic signals at intersections to prioritize emergency vehicles, improving emergency response times while maintaining traffic efficiency. The system employs multimodal detection techniques, including audio classification and object detection, and utilizes LF's deterministic coordination to ensure seamless execution across the computing continuum. We analyze two deployment scenarios: cloud-assisted and fully edge-based. The CAL theorem guides tradeoffs between consistency, availability, and latency, informing optimal service placement at design time. Experimental results validate the theoretical analysis, showing that the edge-based deployment achieves 2.8x lower inference-to-actuation latency and 10.26% lower energy consumption compared to the cloud-assisted scenario, while also eliminating bandwidth overhead associated with data transmission to the cloud.