An Anomaly Detection Framework Based on Data Center Operation and Maintenance Data
Xin Liu, Jibin Wang, Shuwei Qin · 2022
Data centers need to monitor various metrics of their different application platforms and applications in real-time. As the system architectures and application services of different application platforms within them become more complex, the requirements for their anomaly detection capabilities are higher. Therefore, this paper proposes an anomaly detection framework based on data center operation and maintenance data. The framework in this paper consists of three parts, including operation and maintenance data cleaning, data feature extraction, and model routing. It is used to select the appropriate model through model routing based on the indicators such as stability and periodicity obtained from data feature extraction of each application platform. At the same time, in order to enhance the expansion capability of the detection algorithm, a cloud-ground hybrid framework is used and a module for algorithm model management is designed to facilitate interaction with the cloud. After testing on SWAT and WADI datasets, the anomaly detection algorithm with the addition of model routing in the framework has good accuracy and recall performance compared to a single anomaly algorithm model, showing advantages in the task of identifying anomalies.