Smart Energy-Aware Edge Computing for IoT: Real-Time Task Scheduling and Thermal-Aware Load Balancing on Heterogeneous Micro-Clouds

Project Abstract

In this work, we address the challenge of delivering high-performance, energy-efficient, and thermally safe execution for IoT workloads through smart edge computing on heterogeneous micro-clouds. The proposed framework integrates real-time task scheduling, thermal-aware load balancing, and energy optimization across heterogeneous edge resources comprising CPUs, GPUs, and specialized accelerators with diverse power and thermal envelopes. We present a multi-layered scheduling architecture that dynamically classifies tasks by priority, deadline, data locality, and compute intensity, then maps them to the most suitable edge node while considering instantaneous thermal states, coolant efficiency, and power caps. A novel thermal-aware scheduling policy leverages on-device thermal sensors and remote temperature monitoring to predict hot-spots using lightweight machine learning models, enabling proactive throttling and migration to maintain quality-of-service while minimizing thermal throttling and energy waste. The system also introduces an energy-aware task placement strategy that exploits heterogeneous micro-cloud characteristics, including DVFS states, clock gating, and energy-proportional performance curves, to minimize energy per useful computation under given latency constraints. We design an adaptive load-balancing algorithm that distributes workload not only to minimize makespan but also to flatten temperature distribution across the cluster, reducing peak thermals and extending hardware longevity. To validate the approach, we implement a prototype on a testbed consisting of heterogeneous edge nodes with CPUs, GPUs, and AI accelerators, connected via a low-latency, high-bandwidth network. We evaluate under representative IoT scenarios including smart city sensing, industrial automation, and autonomous edge devices, with workloads featuring real-time inference, data fusion, streaming analytics, and periodic control tasks. Our experiments measure latency, energy per inference, thermal margin, and reliability under variable ambient conditions and workload bursts. The results show significant improvements in end-to-end latency and QoS adherence while achieving up to 40–55% reductions in energy consumption during peak load periods and a 20–35% reduction in peak node temperatures compared with baseline static scheduling approaches. Furthermore, the framework demonstrates robust performance under hardware heterogeneity, gracefully handling CPU-GPU-accelerator collaboration, dynamic workload shifts, and thermal throttling events without violating safety margins. We present a case study on a smart transportation scenario where real-time object detection and sensor fusion run concurrently with legacy control loops, highlighting the trade-offs between latency, energy efficiency, and thermal stability. The work contributes a cohesive, cross-layer methodology for energy-aware, thermally conscious, and latency-aware edge computing in IoT ecosystems and provides a scalable blueprint for deployment in future heterogeneous micro-clouds. Finally, we discuss limitations, potential security implications of proactive task migration, and directions for extending the model with predictive cooling, battery-aware edge nodes, and reinforcement learning-based control policies.

Project Overview

What This Project Is About

---

A plain-language overview of how edge devices, like sensors and local servers, work together with cloud micro-clouds to run applications for IoT. The project studies how to schedule tasks in real time on devices with different capabilities while balancing electrical energy use and heat generation so the system runs smoothly without overheating.

---

---

The Problem It Addresses

---

IoT systems at the edge often have varied hardware (different processors, memory) and limited power. This makes it hard to pick which device runs each task without wasting energy or causing slowdowns. Excess heat can throttle performance. The project aims to reduce energy use and prevent overheating while keeping tasks responsive.

---

---

Objectives of the Project

---

1. Understand real-time task requirements and the capabilities of heterogeneous edge devices.
2. Design a simple scheduling strategy that balances speed and energy use.
3. Develop a thermal-aware approach to avoid overheating in micro-clouds.
4. Prototype a lightweight system on a small edge cluster to test decisions.
5. Evaluate energy savings and performance under realistic IoT workloads.

---

---

What You Will Do Step by Step

---

1. Study basic concepts of edge computing, IoT, and energy/thermal management.
2. Set up a small testbed with devices of different capabilities.
3. Implement a simple real-time task scheduler and a thermal monitor.
4. Run workloads to compare energy use and response times with and without the strategy.
5. Analyze results and adjust the scheduling and cooling logic as needed.

---

---

Expected Outcome

A practical, easy-to-implement scheme that reduces energy and heat while maintaining acceptable response times for IoT workloads. The project should produce a report, some simple code, and a proof-of-concept demonstration showing improved efficiency in a heterogeneous edge environment.