Measuring Architectural Misfit: A Preliminary Real Life Study

摘要

Context: A well suited physical structure and architecture is vital for large-scale software development as they impact success of development and the whole project. During development however, a well-designed architecture is vulnerable to subtle decay. Objective: This preliminary study aims for initial indication of this phenomenon for a large-scale software project consisting of hundreds of developers spread all over the world. Information about architectural fit especially over time could be useful for management and software architects. In fact, moves of reengineering could then be anticipated and calibrated to happen regularly. While development would be deferred, the architecture of the software product would be manageable and agility would be preserved. Method: A sliding window approach over the time has been applied. Then for each time window, the prediction error of the duration of an incremental build job estimated by Random Forest models has been measured. Results: As measurements indicate, prediction error increases with advancement in time. Conclusion: Results indicate an increase in misfit.