Multi-tier architecture for Web search engines

We describe a novel multitier architecture far a search engine. Based on observations from query log analysis as well as properties of a ranking formula, we derive a method to tier documents in a search engine. This allows for increased performance while keeping the order of the results returned, and hence relevance, almost "untouched". The architecture and method have been tested large scale on a carrier-class search engine with 1 billion documents. The architecture gives a huge increase in capacity, and is today in use for a major search engine.