Memetic algorithms for the automatic discovery of software architectures

Abstract

Designing complex software systems involves making choices. Particularly in the early stages, software architects need to consider a wide range of design alternatives in order to satisfactorily meet all the requirements. Search techniques like evolutionary algorithms can support them to better explore these choices, requiring the resolution of the formulated search problem under the guidance of software metrics. Aspects like the system complexity and the diversity of metrics have a significant impact on the algorithm performance. Therefore, trying to improve its effectiveness by combining it with local search techniques represents an interesting option to fine tune the returned results. However, designing this kind of hybrid approaches, a.k.a. memetic algorithms, pose new challenges such as determining where these improvements should be located within the evolutionary process. Two memetic approaches are proposed with the aim of analysing and comparing their performance when dealing with a complex problem like the discovery of software architectures.