On Large Execution Traces and Trace Abstraction Techniques

On Large Execution Traces and Trace Abstraction Techniques

Cornelissen, S.G.M.
Moonen, L.

Electrical Engineering, Mathematics and Computer Science

Software Computer Technology

2008-12-31

Preprint of paper published in: Journal of Software Maintenance and Evolution (Wiley), 2008 Program comprehension is an important concern in the context of software maintenance tasks because these activities generally require a certain degree of knowledge of the system at hand. Although the use of dynamic analysis for information gathering has become increasingly popular, the literature indicates that dealing with the excessive amounts of data resulting from dynamic analysis remains a formidable challenge. Although various trace abstraction techniques have been proposed to address these scalability concerns, such techniques are typically not discussed in terms of properties such as complexity and information preservation, and lack thorough evaluation of technique-specific parameters. Moreover, the absence of a common execution trace repository makes matters even worse, as most researchers test their techniques only on their own particular traces. Consequently, it is very difficult to make a fair comparison between the abstraction techniques known from literature. In this paper, we present a characterization of large execution traces in which a set of key properties is extracted from a series of seven representative traces from six different object-oriented systems. Having highlighted the key issues in this context, we propose an assessment methodology for the quantitative evaluation and comparison of trace abstraction techniques. We apply this methodology on a selection of three light-weight abstraction methods, assessing them on the basis of metrics that are relevant in their evaluation.We discuss the results, compare the techniques, and relate the measurements to the trace characteristics found earlier. Our findings provide a valuable insight into factors that play an important role in coping with large execution traces, and we consider our work a significant step towards a systematic assessment of (new) trace abstraction techniques.

program comprehension
dynamic analysis
execution traces

http://resolver.tudelft.nl/uuid:2c4662d7-9251-4835-81bb-062e8b735cbb

Delft University of Technology, Software Engineering Research Group

1872-5392

Technical Report Series TUD-SERG-2008-005

Institutional Repository

report

(c) 2008 The Author(s)
Wiley