Abstract
Combining static and dynamic information is highly relevant in many reverse engineering, program comprehension and maintenance tasks. To allow dynamic information reflecting software system usage scenarios, it should be collected during a long period of time, in a real user environment. This, however, poses several challenges. First and foremost, it is necessary to model the extraction of any relevant dynamic information from execution traces, thus avoiding to collect a large amount of unmanageable data. Second, we need a distributed architecture that allows to collect and compress such an information from geographically distributed users. We propose a probabilistic model for representing dynamic information, as well as a web-service based distributed architecture for its collection and compression. The new architecture has been instantiated to collect inter-procedural program execution traces up to a selectable level of calling context sensitivity. The paper details the role and responsibilities of the architecture components, as well as performance and compression ratios achieved on a set of C and Java programs.