Scaling distributed information systems (DIS) to thousands of information sources raises specific requirements to the discovery of relevant sources for a given application task or query since each source generally covers a very small clipping of the overall domain of discourse only. We propose a novel approach for concisely describing these domain clippings of the sources by defined classes based on a shared ontology. To allow for distributed DIS at a global scale, we present a powerful index structure named SDC-Tree for indexing millions of these source descriptions for retrieving those descriptions that match a given query efficiently.