Customizing the configuration process of an operating system using hierarchy and clustering

Configuration is the process of composing a system from a predefined set of objects. In our approach these objects and their dependencies are described in a graph-based model. Depending upon the requirements of a given application, a configurator chooses required services from a component library by following a set of rules that govern the configurator. When a well-defined hierarchical clustering mechanism for the object graph joins the process preceding the configuration step, the advantages are manifold. The rules of clustering aid the configuration process by not only encapsulating the objects as clusters and sub-clusters but also streamlining the configuration process. The configurator chooses optimized paths by taking advantage of clustered objects, for a given set of reasonably complex requirements. Also, when a change in the set of requirements takes place, it directly attacks the cluster in question with the aid of rules instead of doing the global search for identification of the involved objects. Added advantages result from keeping the rules of clustering inline with the configuration process itself. In this case the entire process is able to support different clusters for different applications. In this paper this approach is discussed in the context of configuration based real-time operating systems