Detecting Anomalies in Server Performance

This thesis studies ways to detect anomalies in server performance and tests simple implementations on detect two different anomalies: CPU related anomalies and memory leaks. The goal is to find a way to implement a lightweight analysis component. This thesis was done as an assignment for Patria Aviation Oy. Common system performance metrics are presented of which CPU and memory utilization are also used later in thesis. A literary study was done to find different strategies on anomaly detection, including both statistical methods and machine learning methods. The simplest strategies, z-score and regression analysis using ordinary least squares, were used for implementation and testing. Testing the implementations showed that z-score was suitable to detecting the anomalies that we were looking for. However, simple linear regression was not robust enough even with smoothed data