A Hyperparameter Tuning Approach for an Online Log Parser

The European Organization for Nuclear Research (CERN) has deployed ALICE'S upgraded computing system in 2020 for improving the performance of the system. One of the aims of the upgraded computing system is to complement the monitoring system by using an Al-based logging system since logs include valuable system runtime information. This allows developers and administrators to monitor their systems and identify abnormal behavior and errors. The new computing system is expected to generate large quantities of logs due to the scale and the complexity of the system. Therefore, log parsing is required to transform unstructured log or free-text log messages into structured logs where the structured logs are ready to use as the input of an automated monitoring system in ALICE. Drain is a popular online log parsing method using the parsing tree technique. However, the performance of Drain depends on the values of parameters (i.e., similarity threshold, maximum depth of the tree, and maximum child nodes of the tree). To achieve the best performance in a reasonable time, we propose a hyperparameter tuning approach by using the Artificial Bee Colony (ABC) algorithm to support Drain. We evaluate our proposed method on two log datasets which are HDFS and Apache ZooKeeper in terms of precision, recall, f-measure, and parsing accuracy.