A Specialized Computer Architecture for Text Retrieval

This paper describes a specialized computer architecture for text retrieval that provides a wide range of query capabilities, without the use of indexes of the material retrieved. A distributed approach is employed, with direct search processors. Each search processor is closely associated with one or more disk drives that store the data to be searched and each consists of a comparator for matching query terms, logic elements to combine query terms, a disk controller and a control minicomputer.The key element of the architecture is the comparator, which can store over 64,000 characters of query terms and compare all of these terms simultaneously with data arriving from a disk continuously at more than one million bytes per second. Exact match, partical match, numeric comparison and context limitation are among the capabilities provided. The comparator is implemented as three similar universal finite-state automata, with special features to facilitate text retrieval and reduce hardware costs.A breadboard version of the system with one searcher is operational and has undergone detailed evaluation. A prototype system with two searchers is being implemented and will be placed into regularly scheduled use.