Mining Partially Periodic Patterns With Unknown Periods From Event Stream

Abstract

Temporal, periodic behavior is common in many application data [6][7][11] including web logs, stock data, alarms of telecommunications, and event logs of computer networks. This is because of routine behaviors of humane (e.g. morning rush every work day), seasonal effects (e.g. increase sales of stationary before a semester), or a consequence of routine tasks (e.g. rebooting print servers every morning or checking stock prices every 10 minutes). An example in computer system is that five repetitions every 10 minutes of a port-down event followed by a port-up event, which in turn is followed by a random gap until the next repetitions of these events. In fact, our study of event logs in production computer networks found that over 50% of the events can be explained by periodic temporal patterns.