Failure proximity

摘要: Recent software systems usually feature an automated failure reporting system, with which a huge number of failing traces are collected every day. In order to prioritize fault diagnosis, due the same expected be grouped together. Previous methods, by hypothesizing that similar imply fault, cluster based on literal trace similarity, we call proximity. However, since can triggered in many ways, quite different. Therefore, previous methods actually group together exhibiting behaviors, like branch coverage, rather than fault. this paper, propose new type proximity, called R-Proximity, regards two as if they suggest roughly location. The location each case suggests is automatically obtained Sober, existing statistical debugging tool. We show addition, find R-Proximity helpful for debugging: It help developers interpret and utilize result. illustrate usage study grep program some experiments Siemens suite, result clearly demonstrates advantage over