A heuristic for legacy information system re-modularization

Abstract

During software evolution, the architecture of the system, which underwent continuous modifications, usually drifts away from its original design and often causes quality declining. Nevertheless, redeveloping systems is risky and expensive and it could take years before the legacy system is replaced by a reliable and equivalent system. Re-modularization is a feasible technique used in automatically restructuring a legacy information system to improve architecture quality and to protect the previous investment. A combinatorial optimization model is proposed to formulate the legacy system re-modularization problem. Unlike previous studies, a new metric of balancing between coupling complexity and cohesion complexity is introduced to avoid the over-cohesive issue. In addition, making use of the problem characteristics, a heuristic based on edge-contraction and vectorization is designed to solve the re-modularization model, and numerous experiments illustrate the fact that the proposed heuristic may lead to better solutions than GA (genetic algorithm) in most cases, and it needs shorter computation times than GA and the heuristic without vectorization.