Effective Segmentation of Large Execution Traces Using Probabilistic and Gaussian Mixture Models

Software maintenance is known to be a costly and time consuming activity. Software engineers need to spend a considerable amount of time in understanding the system before maintaining it. This is due to many reasons including the lack of good documentation and the shift of the original developers of the system to other projects or companies. Dynamic analysis techniques, more particularly trace analysis, are used to alleviate the program comprehension problem by offering software engineers a set of techniques that can help them understand the behavioural aspects of software systems. Execution traces however can be extremely large, which makes them cumbersome for effective analysis. There is a need to develop techniques to help software engineers understand the content of large traces despite their massive size. In this thesis, we present, SumTrace, a novel trace analysis technique. SumTrace takes a trace as input and automatically segments it into smaller and more manageable groups that reflect the execution phases of the traced scenario. The execution phases are summarized to help software engineers understand quickly different parts of the trace without having to analyze its entire content. SumTrace relies on a combination of probabilistic and Gaussian mixture models. We applied SumTrace to the segmentation of large traces, generated from two software systems. The results are very promising. SumTrace is also fast since it only requires only one pass through a trace