A Framework for Failure Diagnosis

Abstract

Testing and debugging is one of the most expensive and challenging phases in the software development life-cycle. One important cost factor in the debugging process is the time required to analyze failures and repair underlying faults. Two types of methods that can help testers to reduce this analysis time are Failure Clustering and Fault Localization. Although there is a plethora of these methods in the literature, there are still some gaps that prevent their operationalization in real world contexts. In addition, the abundance of these methods confuses the practitioners in selecting the suitable method for their own specific domain. To fill the gaps and bring state-of-art closer to practice, we develop a framework for failure diagnosis. To devise this framework, we evaluate existing methods to investigate the possibility of ?nding a method(s) that would be effective in different contexts. Then, we introduce a methodology for adapting this method(s) to different contexts with a priori parameter setting. This framework will empower practitioners to do fast and reliable debugging.