Automatically Generating Predicates and Solutions for Configuration Troubleshooting

Technical support contributes 17% of the total cost of ownership of today’s desktop computers [12], and troubleshooting misconfigurations is a large part of technical support. For information systems, administrative expenses, made up almost entirely of people costs, represent 60–80% of the total cost of ownership [5]. Prior work [21, 24] has created promising tools that automate troubleshooting, thereby saving users time and money. However, these tools assume the existence of predicates, which are test cases painstakingly written by an expert. Since both experts and time are in short supply, obtaining predicates is difficult. In this paper, we propose a new method of creating predicates that infers predicates by observing the actions of ordinary users troubleshooting misconfigurations. We report on the results of a user study that evaluates our proposed method. The main results were: (1) our method inferred predicates for all configuration bugs studied with very few false positives, (2) using multiple traces improved results, and, surprisingly, (3) our method identified the correct solutions applied by users who fixed the misconfigurations in the study.