Rapid Trend Prediction for Large-Scale Cloud Database KPIs by Clustering
For cloud database manufacturers, it is an essential work to monitor a large number of KPIs (Key Performance Indicators) for each database instance and ensure their service quality. To provide intelligence scalability of cloud database, KPIs trend prediction has been proposed to guide operation and maintenance team to adjust cloud resources reasonably and timely. Existing KPIs trend prediction usually build prediction model for each KPI, and it is not easy to be widely applied due to massive resource consumption. In this paper, we propose a rapid KPI trend prediction framework TPC(Trend Prediction based on Clustering). It consists of four steps: preprocessing original KPIs streaming data, clustering KPIs based on the shape similarity, building trend prediction model for each cluster centroid, and predicting a new KPI with the prediction model of its cluster centroid. During clustering KPIs, we improve a state-of-the art clustering algorithm ROCKA by finding a better optimized density radius. The evaluation experiments are conducted on three public and two industrial dataset, and the results indicate that our improved ROCKA could cluster KPIs with higher accuracy. Moreover, the experiments on two industrial dataset show that TPC could reduce much more training time with less prediction performance loss.
Sat 29 MayDisplayed time zone: Amsterdam, Berlin, Bern, Rome, Stockholm, Vienna change
14:15 - 15:00
|Robust and Transferable Anomaly Detection in Log Data using Pre-Trained Language Models|
|Rapid Trend Prediction for Large-Scale Cloud Database KPIs by Clustering|
|Learning Dependencies in Distributed Cloud Applications to Identify and Localize Anomalies|