Parallel computation of the Burrows Wheeler Transform in compact space

Abstract

The Burrows-Wheeler Transform (BWT) has become since its introduction a key tool for representing large text collections in compressed space while supporting indexed searching: on a text of length n over an alphabet of size σ, it requires O(nlg⁡σ) bits of space, instead of the O(nlg⁡n) bits required by classical indexes. A challenge for its adoption is to build it within O(nlg⁡σ) bits as well. There are some sequential algorithms building it within this space, but no such a parallel algorithm. In this paper we present a PRAM CREW algorithm to build the BWT using O(nlg⁡n) work, O(lg3⁡n/lg⁡σ) depth, and O(nlg⁡σ) bits.