Simulated Annealing for Aiding Genetic Algorithm in Software Architecture Synthesis

AbstractThe dream of software engineers is to be able to automatically producesoftware systems based on their requirements. Automatic synthesis of soft-ware architecture has already been shown to be feasible with genetic algo-rithms. Genetic algorithms, however, easily become very slow if the size ofthe problem and complexity of mutations increase as GAs handle a large pop-ulation with much data. Also, for purely scientific interest it is worthwhileto investigate how other search algorithms handle the problem of softwarearchitecture synthesis. The present paper studies the possibilities of usingsimulated annealing for synthesizing software architecture. For this purposewe have two goals: 1) to study whether a simpler search algorithm can handlesynthesis and 2) if a seeded algorithm can provide quality results faster than asimple genetic algorithm. We start from functional requirements which forma base architecture and consider three quality attributes, modifiability, effi-ciency and complexity. Synthesis is performed by adding design patterns andarchitecture styles to the base architecture. The algorithm thus produces asoftware architecture which fulfills the functional requirements and also corre-sponds to the quality requirements. It is concluded that simulated annealingas such does not produce good architectures, but it is useful for speeding upthe evolution process by quickly fine-tuning a seed solution achieved with agenetic algorithm. The main contribution is thus a new seeded algorithm forsoftware architecture design.Keywords: search-based software engineering, simulated annealing, softwaredesign, genetic algorithm, software architecture