Design and Implementation of EMKOS for Edge Computing

As embedded systems increasingly migrate to- wards the edge, handling complex and intermittent workloads has become a critical challenge. Traditional static scheduling algorithms, such as Round Robin or Fixed Priority, often fail to provide optimal response times in these dynamic environments where task characteristics fluctuate wildly. This paper presents the design and implementation of EMKOS, a custom micro- kernel architecture specifically tailored for Edge Computing applications. We integrate Artificial Intelligence at both the kernel and edge levels to optimize task scheduling and resource allocation. At the kernel level, we propose a “Prediction Enhanced Dynamic Randomized Multi-Level Feedback” (PE- DRMLF) algorithm, which utilizes an Exponential Weighted Moving Average (EWMA) model to dynamically predict task burst times and preemptively assign priorities. Complementing this, an external ESP32-based Edge AI module employs a TinyML Neural Network to classify physical sensor events before they reach the kernel, acting as an intelligent interrupt controller. The system is validated through a novel Hardware- in-the-Loop (HIL) test harness, bridging a physical ESP32 job generator with a QEMU-emulated kernel. Experimental results demonstrate that this hybrid approach significantly reduces response latency for interactive tasks compared to standard scheduling baselines, proving the viability of statistical learning in real-time operating systems.