Testing BFT with Containerized IoT Nodes and Configurable Fault Injection

IoT systems rely on data collected from numerous distributed sensors, which may fail or become compromised. Sensors that exhibit arbitrary or malicious behavior are said to suffer from Byzantine faults, and Byzantine Fault Tolerance (BFT) mechanisms enable the system to continue operating correctly despite their adverse effects. This paper presents a scalable testbed for evaluating Byzantine node detection during IoT sensor fusion. The testbed uses real containerized services that communicate over an MQTT broker, while the sensor signals themselves are simulated. A chaos engine injects various fault types according to a configurable scenario file, a consolidator estimates the correct sensor value despite corrupted readings, and an anomaly detector operates alongside it to flag suspicious data. We compare five statistical consolidation strategies: mean, trimmed mean, median, interquartile mean, and trimean. Experimental results show that robust statistical estimators tolerate compromised sensors well beyond the classical consensus bound of n ≥ 3f+1, remaining effective with up to half of the participating sensors compromised.