Managing trace data volume through a heuristical clustering process based on event execution frequency

To regain architectural insight into a program using dynamic analysis, one of the major stumbling blocks remains the large amount of trace data collected. Therefore, this paper proposes a heuristic which divides the trace data into recurring event clusters. To compose such clusters the Euclidian distance is used as a dissimilarity measure on the frequencies of the events. Manual inspection of these event sequences revealed that the heuristic provides interesting starting points for further examination.