Capturing Expert Knowledge for Automated Configuration Fault Diagnosis

Abstract

The process of manually diagnosing a software misconfiguration problem is time consuming. Manually writing and updating rules to detect future problems is still the state of the practice. Consequently, there is a need for increased automation. In this paper, we propose a three-phase framework using machine learning techniques for automated configuration faults diagnosis. This system can also help in capturing expert knowledge of configuration troubleshooting. Our experiments on Apache web server configurations are generally encouraging and non-experts can use this system to diagnose misconfigurations effectively.