Anomaly detection in the context of long-term cloud resource usage planning.

This paper describes a new approach to automatic long-term cloud resource usage planning with a novel hybrid anomaly detection mechanism. It analyzes existing anomaly detection solutions, possible improvements and the impact on the accuracy of resource usage planning. The proposed anomaly detection solution is an important part of the research, since it allows greater accuracy to be achieved in the long term. The proposed approach dynamically adjusts reservation plans in order to reduce the unnecessary load on resources and prevent the cloud from running out of them. The predictions are based on cloud analysis conducted using machine learning algorithms, which made it possible to reduce costs by about 50%. The solution was evaluated on real-life data from over 1700 virtual machines. [ABSTRACT FROM AUTHOR]