Special-purpose hardware for text searching: Past experience, future potential

摘要

The partitioned finite state automaton (PFSA) was proposed ten years ago as a new way of implementing a hardware-based pattern matcher for text information retrieval systems. The pattern matcher was part of a special-purpose search processor to be attached to each disk of a large information retrieval system. Since then, there have been dramatic changes in microprocessor, memory, and disk technologies. After an overview of the experiences in implementing and evaluating the PFSA searcher, the applicability of special-purpose hardware for text searching in the future is discussed. The use of a surrogate, such as a partially inverted file, in conjunction with the search hardware, is examined, particularly regarding the overhead of the surrogate and any changes to the operation and performance of the search system. Searching using the higher speed RISC microprocessors available today as an alternative to a custom searcher is presented, and a hardware-augmented approach is proposed.