Verifying design modularity, hierarchy, and interaction locality using data clustering techniques

Modularity, hierarchy, and interaction locality are general approaches to reducing the complexity of any large system. A widely used principle in achieving these goals in designing software systems is striving for high cohesion within a module and low coupling between modules. However, this principle has difficulties in practice. Because a hierarchical system structure often consists of several layers, it is difficult to decide at what layer an interaction should be considered as cohesion, and at what layer an interaction should be considered as coupling. In this paper, we do not differentiate cohesion and coupling, but use a general term interaction to represent the dependencies between software modules. We propose a method to verify the design modularity, hierarchy, and interaction locality of a software system. This approach is based on the component interactions gathered from certain design level artifacts, such as UML diagrams. Data clustering technique is then used to group software components according to the degree of interactions between them. To show how to use this approach, we apply it to Parna's KWIC object-oriented design example, in which sequence diagram is used to derive the degree of component interactions.