ThermoLoad: A Multi-Criteria Load Balancing Algorithm for Sustainable and Carbon-Efficient Data Centers

📘 Overview

ThermoLoad is a novel load balancing algorithm designed to optimize server workload distribution based on multiple performance and environmental metrics—primarily server temperature, CPU usage, response time, and hardware weight.

Unlike traditional algorithms (e.g., Round Robin, Least Connection), ThermoLoad is thermally aware and multi-metric, aiming to enhance data center sustainability by minimizing hotspots, reducing response time, and improving energy efficiency.

This repository includes:

• ✅ Implementation of ThermoLoad and five benchmark algorithms
• 🧪 A simulated test environment using realistic server conditions
• 📊 Plots and comparison metrics
• 📄 LaTeX source for the research paper (optional)

---

🧠 Core Idea

Traditional load balancers ignore real-time server thermals and dynamic performance states. ThermoLoad addresses this gap using a weighted scoring formula:

$$\text{Score} = \alpha \cdot T + \beta \cdot C + \gamma \cdot R + \delta \cdot W$$

Where:

• T = Normalized temperature (0 to 1)
• C = Normalized active connections
• R = Normalized response time
• W = Normalized server weight (higher = better)
• α, β, γ, δ = Weighting coefficients for each factor

The server with the lowest score is selected to handle incoming requests, ensuring performance and thermal stability.

⚙️ Experimental Setup

• Language: Python 3.11
• Packages Used: numpy, pandas, matplotlib, random

System Configuration

• CPU: Intel Core i7-12700H @ 2.30GHz
• RAM: 32 GB
• OS: Ubuntu 22.04 LTS

Server Simulation

Parameter	Range
Virtual Servers	6
CPU Frequencies	2.2 – 3.4 GHz
Temperatures	32°C – 61°C
Active Connections	20 – 200
Response Times	15 ms – 150 ms
Server Weights	1 – 5

Workloads Emulated

• CPU-bound (40%)
• Memory-bound (30%)
• I/O-heavy (30%)

---

🧪 How to Run the Code

Install Dependencies

pip install numpy pandas matplotlib

📚 Cited Works and Influences

This project builds upon and extends ideas from notable works, including:

• Hassel, W.J. — Carbon cost of cooling AI data centers (2023)
• Singhal, A. — AI-enhanced load balancing (2023)
• Fan, X. — Power provisioning in warehouse-scale data centers (2007)
• Iffländer, L. — Heat-aware routing models (2022)

See the research paper in the /report directory for full citations.

---

📜 License

This project is intended for academic and research use only.
Please cite the associated paper if used in publications.

---

🤝 Acknowledgments

Special thanks to the open-source community and the authors of prior work in:

• 🔥 Thermal-aware computing
• ☁️ Cloud performance optimization
• 🌱 Sustainable server architecture