Mining Temporal Rules from Program Execution Traces

• Main Authors: LO, David, KHOO, Siau-Cheng, LIU, Chao
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2007
• Subjects: Software Engineering
• Online Access: https://ink.library.smu.edu.sg/sis_research/949; http://portal.acm.org/citation.cfm?id=1401838
• Institution: Singapore Management University

Specification mining is a process of extracting specifications, often from program execution traces. These specifications can in turn be used to aid program understanding, monitoring and verification. There are a number of dynamic-analysis-based specification mining tools in the literature, however none so far extract past time temporal expressions in the form of rules stating: "whenever a series of events occurs, previously another series of events has happened". Rules of this format are commonly found in practice and useful for various purposes. Most rule-based specification mining tools only mine future-time temporal expression. Many past-time temporal rules like "whenever a resource is used, it was allocated before" are asymmetric as the other direction does not holds. Hence, there is a need to mine past-time temporal rules. In this paper, we describe an approach to mine significant rules of the above format occurring above a certain statistical thresholds from program execution traces. The approach start from a set of traces, each being a sequence of events (i.e., method invocations) and resulting in a set of significant rules obeying minimum thresholds of support and confidence. A rule compaction mechanism is employed to reduce the number of reported rules significantly. Experiments on traces of JBoss Application Server shows the utility of our approach in inferring interesting past-time temporal rules.