The subsystem decomposition of a software system degrades gradually during its lifetime and therefore it gets harder and harder to maintain. As a result this decomposition needs to be reconditioned from time to time. The problem is to determine a suitable subsystem decomposition that can be used as a basis for future maintenance tasks. This paper describes a new methodology that computes such a subsystem decomposition by optimizing metrics and heuristics of good subsystem design. The main idea is to treat this task as a search problem and to solve it using a genetic algorithm.

The recovery of software architecture is a first important step towards re-engineering a software system. Architecture recovery usually involves clustering. The problem with current clustering techniques is that they decide exclusively based on syntactic dependencies instead of looking at higher-level semantic information. As a result, the recovered architecture is not always meaningful to a human software engineer.In this paper, we propose an approach that combines clustering with pattern-matching techniques to recover meaningful decompositions. Pattern-matching is used to identify architectural clues -small structural patterns that provide semantic information to allow for a rating of the dependencies found between a system's entities. These clues are used to compute an adaptive inter-class similarity measure which is then used by a clustering algorithm to produce the final system decomposition.