An architecture for parallel search of large, full-text databases

A novel signature approach is introduced for retrieval of documents from large text databases. The structure of the signature files under this approach lends itself to a highly parallel hardware implementation. The signature files are generated by software and loaded into RAM. Upon a user query, the signature files are searched in parallel by a special-purpose search module for identification of relevant documents. The search module is implemented in hardware with a parallel architecture. Experimental results and the design of the search module are presented. Experiments have shown that this technique is quite efficient in terms of storage overhead and the number of false drops. Perhaps the most attractive feature of this system is the hardware cost. The signature files are designed so that they can be saved, accessed, and compared using inexpensive memory chips. Another feature of this system is the capability for simple manipulation of the `don´t care characters,´ which is not provided by other signature methods