Predicting Subsystem Failures using Dependency Graph Complexities

In any software project, developers need to be aware of existing dependencies and how they affect their system. We investigated the architecture and dependencies of Windows Server 2003 to show how to use the complexity of a subsystem´s dependency graph to predict the number of failures at statistically significant levels. Such estimations can help to allocate software quality resources to the parts of a product that need it most, and as early as possible.