Anomaly Detection in Cloud Components

Today’s smart connected world is all about devices, data and connectivity. Cloud computing has been introduced to support the rapidly growing technological demand by providing flexibility, efficiency, mobility, robustness and disaster recovery mechanism to individuals and businesses. Cloud platforms, under the hood, consist of a complex inter-connected stack of hardware and software components. Each of these components can fail, which may lead to an outage. Therefore, anomaly detection accuracy plays a vital role in effective system operation management, which is the key to ensure stability and performance. Our goal is to improve the quality of Cloud services through early detection of such failures by analyzing resource utilization metrics. In this thesis, we constructed a Gated-Recurrent-Unit-based autoencoder to generate reconstruction errors and computed the likelihood of anomalies based on the distribution of reconstruction errors. The combined approach enables efficient detection of abnormalities in multi-dimensional telemetry time-series gathered from a private Cloud.