Aberrant Behavior Detection in Time Series for Network Service Monitoring

The open-source software RRDtool and Cricket provide a solution to the problem of collecting, storing, and visualizing service network time series data for the real-time monitoring task. However, simultaneously monitoring all service network time series of interest is an impossible task even for the accomplished network technician. The solution is to integrate a mathematical model for automatic aberrant behavior detection in time series into the monitoring software. While there are many such models one might choose, the primary goal should be a model compatible with real-time monitoring. At WebTV, the solution was to integrate a model based on exponential smoothing and Holt-Winters forecasting into the Cricket/RRDtool architecture. While perhaps not optimal, this solution is flexible, efficient, and effective as a tool for automatic aberrant behavior detection.