Understanding the impact of IoT security patterns on CPU usage and energy consumption: a dynamic approach for selecting patterns with deep reinforcement learning

Saeid Jamshidi, Amin Nikanjam, Kawser Wazed Nafi, Foutse Khomh

Abstract: The Internet of Things (IoT) introduces numerous security challenges that require effective solutions. IoT security patterns provide a practical approach to address recurring security issues, yet their impact on edge gateway metrics, e.g., energy consumption, CPU usage, and load, remains largely unexplored. This study empirically evaluates six IoT security patterns (Personal Zone Hub, trusted communication partner, outbound-only connection, blacklist, whitelist, and secure sensor node) in three IoT-edge applications: smart home, smart city, and healthcare. We evaluated the patterns individually and in combination, subjecting them to cyber threats. Subsequently, we analyzed their impact on energy consumption, CPU usage, and load. To address observed resource trade-offs, we also propose a deep reinforcement learning-based intrusion detection system that dynamically selects security patterns based on real-time conditions. Tested against threats (for example, DoS Hulk, Slowloris, DDoS, and GoldenEye), this adaptive approach optimizes security and resource efficiency, selecting the most suitable patterns for each scenario. The findings show that pattern selection significantly impacts resource metrics and that the DRL-based system maintains robust security while minimizing energy and CPU overheads. Based on these findings, we provide guidelines for developers to improve IoT-edge security by optimizing resource consumption.