Abstract

This paper presents Ingrid, an architecture for a fully distributed, fully self-configuring information navigation infrastructure that is designed to scale to global proportions. Unlike current designs, Ingrid is not a hierarchy of large index servers. Rather, links are automatically placed between individual resources based on their topic similarity in such a way that clusters of term combinations are formed. The resulting topology can potentially be searched and browsed by a robot efficiently. This paper describes the fundamentals of Ingrid--the topology design and the algorithms for creating and searching the topology. It discusses the scaling characteristics of Ingrid, and gives the scaling results of a limited experiment. Keywords:

Citations