Abstract

Parallelism is useful in the storage and access of structured documents. Fast parallel algorithms for search in structured text are already known, but they will not supplant the use of indexes to speed up searching until massively parallel architectures become routinely available. However, parallel algorithms suggest new kinds of indexes that provide powerful search capability and performance even on modestly-parallel computers. We present a generalisation of indexes based on regular languages, called indexing languages, that are chosen to be homomorphic images of languages generated by typical search patterns. Precomputing properties of text strings relative to indexing languages makes it fast to exclude large parts of the text from consideration before executing a direct search. 1 Background Search times in document archives are improved by building indexes giving the location of (usually) words as offsets in the structure. This kind of index has three major drawbacks: it is expens...

Citations