Fault location method of complex software based on community mining

Abstract

The explosion by amount of codes as well as the swelling logic complexity have stifled the performance of the traditional fault-location methods since the resource adopted during this process is unacceptable. Under such a situation, a scheme to locate the faults in complex software more effectively has been proposed in this paper based on networks community theory. First, on the base of establishing the dependency network of every module in software, dependency relationships between various community blocks can be obtained by applying clustering algorithms to mine the community structure. Second, with status reports generated by monitoring-codes planted into the center node of every community, we can distinguish whether communities work in a normal way, thus the fault searching area can be shrunk smaller by repeating the steps above. Third, once the searching area has been shrunk to several modules, the traditional methods like Tarantula is eligible for finding the fault since the number of suspected codes has been compressed to an acceptable level. At last, the experiment by applying the method we come up has been conducted and the statistic has been collected as well as analyzed to compare with the main-trend solutions, during which the speed superiorities and acceptable accuracy of our method have been confirmed.