Inter-Package Dependency Networks in Open-Source Software.pdf

Inter-Package Dependency Networks in Open-Source Software Nathan LaBelle and Eugene Wallingford Department of Computer Science University of Northern Iowa Cedar Falls, Iowa 50613 labelle@ Abstract The new subject of network science examines large graphs that model real-world, growing, and dynamic systems of interacting components. These networks form outside the realm of centralized control, but exhibit global organization and share common non- trivial properties. This research analyzes networks in Open-Source Software (OSS) at the package level, where dependencies link diverse software systems constructed by otherwise disjoint development groups. Data was mined from two OSS repositories and interaction graphs were constructed. We show that this module coupling creates a clustered network with low separation between packages (a “Small-World”). We show that the distribution of edges in the graph is self- similar at all scales and skewed, yielding a power-law distribution. These properties indicate that package networks share natural organization patterns with previously studied systems in natural and social sciences. We argue that software networks are governed by similar dynamics. Finally, we discuss the implications of network structure on software engineering. 2 Introduction Since Euler’s 1735 solution to the K?nigsberg Bridges problem, a huge amount of knowledge has been built on the topic of mathematical graphs. In the last 5 years alone, at least 21,000 papers have been published on the topic of complex networks (National Academies, 2005). A network is a typically unweighted and simple large graph ( , )G V E= where V denotes a vertex set and E an edge set. Vertices represent discrete objects in a system, such as social actors, economic agents, computer programs, or biological producers and consumers. Edges represent interactions among these “interactons”. For example, if software objects are represented as vertices,