mtjikuzu/carbon-aware-cybersecurity-scheduling

Carbon-Aware Cybersecurity Scheduling (CACS)

Reducing the Carbon Footprint of Security Operations Through Intelligent Workload Deferral

---

🎯 Overview

This repository contains the complete research prototype for Carbon-Aware Cybersecurity Scheduling (CACS) — a novel framework that reduces the carbon footprint of security operations by intelligently deferring non-critical tasks to time windows when the electricity grid has lower carbon intensity.

Key Finding: Carbon-aware scheduling achieves a 28.8% reduction in CO₂ emissions from security operations with only 3.8 hours average additional vulnerability exposure for deferrable tasks.

📊 Results Summary

Strategy	CO₂ (kg)	Reduction	Avg Exposure (h)	Max Exposure (h)
Fixed (Baseline)	98.52	—	0.0	0
Carbon-Aware	70.10	28.8%	3.8	162
Hybrid (α=0.6)	78.77	20.0%	1.1	22
Night-Shift	81.65	17.1%	1.6	18

Statistical robustness: 29.5% ± 1.2% mean reduction (95% CI: 27.1–31.8%) over 10 independent trials.

📁 Repository Structure

├── README.md                    # This file
├── METHODOLOGY.md               # Full experimental design document
├── experiment.py                # Complete working prototype (run this!)
├── paper_draft.tex              # Springer ACSAR formatted paper draft
├── figures/
│   ├── fig1_carbon_intensity_pattern.png
│   ├── fig2_strategy_comparison.png
│   ├── fig3_pareto_front.png
│   ├── fig4_savings_by_task_type.png
│   └── fig5_scheduling_timeline.png
└── results/
    ├── strategy_comparison.csv
    └── task_type_analysis.csv

🚀 Quick Start

# Install dependencies
pip install numpy pandas matplotlib

# Run the full experiment
python experiment.py

# For real UK carbon intensity data (requires internet):
# Edit experiment.py: set use_real_data=True

📐 Problem Formulation

Given a set of security tasks T = {t₁, t₂, ..., tₙ}, each with energy consumption eᵢ, generation time gᵢ, deadline dᵢ, and deferrability flag δᵢ, minimize:

min Σ eᵢ · c(sᵢ)

Subject to:

• sᵢ = gᵢ if δᵢ = 0 (non-deferrable)
• gᵢ ≤ sᵢ ≤ gᵢ + dᵢ if δᵢ = 1 (deferrable)

Where c(h) is the carbon intensity (gCO₂/kWh) at hour h.

🔬 Scheduling Strategies

1. Carbon-Aware Greedy — Defer each task to the lowest-carbon hour within its deadline
2. Hybrid (α-parameterized) — Balance carbon reduction vs. urgency: score(h) = α·ĉ(h) + (1-α)·u(h)
3. Night-Shift Heuristic — Defer to 00:00-06:00 (industry baseline)
4. Fixed Schedule — Execute immediately (baseline)

📈 Data Sources

Data	Source	Access
Carbon Intensity (UK)	api.carbonintensity.org.uk	Free, no auth
Carbon Intensity (Historical)	electricityinfo.org	Free
Security workload model	Synthetic (SANS SOC Survey-based)	Generated in code

📝 Conference Submission

• Conference: ICISET 2026 — Namibia University of Science and Technology
• Track: Infrastructure, Security and Governance — Cybersecurity and Sustainability
• Format: Springer ACSAR Series (LNCS style)
• Deadline: July 27, 2026

📚 Key References

1. Lannelongue et al. "Green Algorithms: Quantifying the carbon footprint of computation" (2021)
2. Lin et al. "Chasing Low-Carbon Electricity for Practical and Sustainable DNN Training" (2023)
3. UK National Grid Carbon Intensity API
4. NIST Cybersecurity Framework v2.0 (2024)

🏗️ Future Work

• Real-world SOC deployment with direct energy measurement
• Carbon intensity forecasting for predictive scheduling
• Multi-region scheduling leveraging geographic carbon differences
• Reinforcement learning adaptive policies
• Namibia/Southern Africa carbon intensity modeling

📄 License

MIT License — feel free to use, modify, and extend for your research.

Generated by ML Intern

This model repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

• Try ML Intern: https://smolagents-ml-intern.hf.space
• Source code: https://github.com/huggingface/ml-intern

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "mtjikuzu/carbon-aware-cybersecurity-scheduling"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

For non-causal architectures, replace AutoModelForCausalLM with the appropriate AutoModel class.