Anomaly Detection and Analysis for Reliability Management Clustered Container Architectures

Virtualised environments such as cloud and edgecomputing architectures allow software to be deployed andmanaged through third-party provided services. Here virtualisedresources available can be adjusted, even dynamically to changingneeds. However, the problem is often the boundary between theservice provider and the service consumer. Often there is no directaccess to execution parameters at resource level on the provider'sside. Generally, only some quality factors can be directly observedwhile others remain hidden from the consumer. We propose anarchitecture for autonomous anomaly analysis for clustered cloudor edge resources. The key contribution is that the architecturedetermines possible causes of consumer-observed anomalies inan underlying provider-controlled infrastructure. We use HiddenHierarchical Markov Models to map observed performanceanomalies to hidden resources, and to identify the root causes ofthe observed anomalies in order to improve reliability. We applythe model to clustered hierarchically organised cloud computingresources. We illustrate use cases in the context of containertechnologies to show the utility of the proposed architecture