Detecting and Predicting Anomalies for Edge Cluster Environments using Hidden Markov Models

Edge cloud environments are often build as virtualized coordinated clusters of possibly heterogeneous devices. Their problem is that infrastructure metrics are only partially available and observable performance needs to be linked to underlying infrastructure problems in case of observed anomalies in order to remedy problems effectively. This paper presents an anomaly detection and prediction model based on Hidden Markov Model (HMM) that addresses the problem of mapping observations to underlying infrastructure problems. The model aims at detecting anomalies but also predicting them at runtime in order to optimize system availability and performance. The model detects changes in response time based on their resource utilization. We target a cluster architecture for edge computing where applications are deployed in the form of lightweight containers. To evaluate the proposed model, experiments were conducted considering CPU utilization, response time, and throughput as metrics. The results show that our HMM detection and prediction performs well and achieves accurate fault prediction.