Fault-aware Fingerprinting: Towards Mutualism between Failure Investigation and Statistical Debugging

Failure investigation and statistical debugging are two critical components in automated failure analysis. Although related, the two components have always been studied separately in previous work. In this paper, we propose a faultaware fingerprinting technique that relates the two components and puts them into mutualism, i.e., each component benefits and benefits from the other. Technically, we use statistical debugging algorithms to find the fault location each failing trace suggests, and take the suggested fault location as the fingerprint of the failing trace. The fingerprint is fault-aware in that it embodies the fault location each failing trace suggests. With fingerprinting, we can cluster together failing traces that suggest similar fault locations, rather than those that are literally similar to each other as done by previous work. We observe from experiments that the fingerprint-based clustering renders more meaningful results than the literal trace similarity-based approach. On the other hand, we also observe that the investigation of failing traces in the fingerprint space can help developers interpret statistical debugging results. In this paper, we report on the achieved results that exemplify the promises of the fault-aware fingerprinting technique, and propose future work for discussion and comments. Advisor: Dr. Jiawei Han. [email protected].