Network Conditions Dominate MQTT-over-TLS 1.3 Performance: A Comparative Evaluation of ML-KEM-768, RSA-2048, and ECC-P256

The expected arrival of cryptographically relevant quantum computers requires Internet of Things (IoT) communication infrastructure to migrate to Post-Quantum Cryptography (PQC). However, it remains unclear how this migration affects protocol-level performance of IoT communication under controlled network degradation rather than algorithm cost alone. We used the PQC-IoTNet SITL framework to benchmark ML-KEM-768, the NIST-standardized post-quantum Key Encapsulation Mechanism, against RSA-2048 and ECC-P256 in MQTT-over-TLS 1.3 IoT-to-cloud communication across 20 network scenarios (five latency levels × four packet loss levels) and three algorithms (60 algorithm network configurations, 3000 trials). The principal finding is that network conditions, rather than cryptographic algorithm choice, dominate protocol-level performance. Under normal network conditions, ML-KEM-768 showed handshake performance comparable to the classical algorithms, while under degraded conditions, algorithm-level differences remained small relative to network-driven variance. Also, ML-KEM-768 showed an approximately 2.5 ms CPU-time advantage under congestion and an absolute peak Python-process RSS within 0.1 MB of the classical configurations, at the cost of 6.7% more bytes transmitted than RSA-2048 (1.3% fewer than ECC-P256). Connection reliability was 99.8% for ML-KEM-768 compared with 100% for RSA-2048 and ECC-P256. The only two unsuccessful connection attempts observed in the entire study occurred with ML-KEM-768 under high-impairment conditions. However, Fisher’s exact test indicated that this difference was not statistically significant at the present sample size (p = 0.50). These findings, within the evaluated SITL environment, indicate that ML-KEM-768 delivers protocol-level performance comparable to RSA-2048 and ECC-P256 under the tested network conditions while providing a standardized post-quantum key establishment mechanism for MQTT-based IoT communication. Further validation on resource-constrained hardware is required before generalizing these results to production IoT deployments.