Enhancing Reasoning Approaches to Diagnose Functional and Non-Functional Errors

Most approaches to automatic software diagnosis abstract the system under analysis in terms of component activity and correct/incorrect behaviour (colectivelly known as spectra). While this binary error abstraction has been shown to be capable of diagnosing functional errors, when diagnosing non-functional errors it yields suboptimal accuracy. The main reason for this limitation is related to the lack of mechanisms for encoding error symptoms (such as performance degradation) in such a binary schema. In this paper, we propose a novel approach to diagnose both functional and non-functional errors by incorporating into the classic, bayesian reasoning approaches to error diagnosis concepts from the fuzzy logic domain. The empirical evaluation on 27000 synthetic scenarios demonstrates that the proposed fuzzy logic-based approach considerably improves the diagnostic accuracy (20% on average, with 99% statistical significance) when compared to the classic, state-of-the-art approach.