doi:10.1093/comjnl/bxv049

Software architecture design is a complex task, requiring handling and balancing multiple tradeoffs. In this paper, the potential of genetic algorithms (GAs) in automated software architecture design is explored, assuming that software architecture is constructed of patterns. We have implemented different techniques based on GAs with variations on algorithmic operations and evaluation func-tions. We perform an extensive case study using a real framework system as a benchmark. The solu-tions are analyzed and compared with the man-made design of the framework. Our purpose is to study what kind of pattern configurations the algorithm is able to produce, how close they are to the ones used by a human and whether modifying the algorithm gives better solutions. Results show that 60 % of the patterns proposed by the algorithm can be seen as well-placed, but there are big differ-ences between the techniques and certain patterns are significantly more difficult for the algorithm to handle than others