Software Fault Localization Based on Centrality Measures

The existence of software faults is the cause of software failure. Locating the faults hidden behind software is a central task towards efficient software development and maintenance. Many techniques have been developed to help locate software faults. However, they have some obvious shortcomings. Inspired by the surprisingly discovery in the crossover studies of software and complex networks, in this paper, a fault localization method based on software network centrality measures (SNCM) is proposed. Software execution traces are modeled as networks and two centrality measures (node degree and structural hole) are adopted to calculate the suspiciousness of each statement. Modeling software from a network view can help handle the increasing complexity involved in fault localization. Moreover, the software network measures are essential because they focus on the macro-level software statistics. The universality of the proposed method is thus promising. Statistically repeatable experiment is applied to the programs in the Siemens suite and the results demonstrate the effectiveness of the proposed method and the poor performance of the Tarantula method under random inputs.