Shells Bells: Cyber-Physical Anomaly Detection in Data Centers

Monitoring the side-channel sound can improve anomaly detection (AD) in data centers (DCs). However, a DC’s dense setup results in a composite soundscape which makes it difficult to attribute sounds to individual devices. We propose a novel cyber-physical AD approach that validates device activity in realistic composite audio signals. By leveraging information from management network traffic, we predict changes in the DC soundscape. We use a convolutional neural network to compare our predictions with real observations to
validate correct device activity and identify anomalies. Our evaluation using data from a real DC environment identifies spoofed and masqueraded activity with an accuracy of 98.62%.