A study is made of the design characteristics of two very different approaches to performing searches within the context of text based information retrieval. The first represents a recent advance to the centralized index search engine. Here most emphasis is placed on the efficient representation of the keyword indexes, as used to express the association between documents and keywords of the global search schema, without loosing retrieval effectiveness. On the other hand a recently proposed, completely decentralized, index based search methodology is reviewed. This does not employ a search engine as such, but relies on controlling the propagation of search agents across the network of heterogeneous information providers. In this case the physical allocation of information to the relevant information providers in accordance with a virtual tree organization is significant.