Time-Series Analysis for Performance Monitoring and Anomaly Detection in Computer Networks
Marcos Portnoi, Priscilla Moraes, Martin Swany · 2010
We survey, in this work, applicationsfor time‐series statistical analysis of computer network data, specifically for performanceandanomalydetection. In therealmofQuality of Service, network agents could control the fair distribution ofresourcesbasedonhistoricalbehaviorofapplications, instead of on deterministic algorithms. Virtual circuits, for instance, can be allocated on demand for applications that exhibit a past of high utilization. Furthermore, in a network performance monitoring architecture, such as perfSONAR, services may benefit from time‐series analysis of measurement data to trigger events, audit statistical behavior, or detectanomaliesinthenetwork. Theseanomaliesmight indicateperformanceorsecurityissues. Finally, time‐series analysis enables forecasting, that can be employed to predict future performance.