Hierarchical Small Worlds in Software Architecture

In this paper, we present a complex network approach to the study of software engineering. We have found universal network patterns in a large collection of object-oriented (OO) software systems written in C++ and Java. All the systems analyzed here display the small-world behavior, that is, the average distance between any pair of classes is very small even when coupling is low and cohesion is high. In addition, the structure of OO software is a very heterogeneous network characterized by a degree distribution following a power-law with similar exponents. We have investigated the origin of these universal patterns. Our study suggests that some features of OO programing languages, like encapsulation, seem to be largely responsible for the small-world behavior. On the other hand, software heterogeneity is largely independent of the purpose and objectives of the particular system under study and appears to be related to a pattern of constrained growth. A number of software engineering topics may benefit from the present approach, including empirical software measurement and program comprehension.