Parallel approaches to permutation-based indexing using inverted files

Abstract

We present parallel strategies for indexing and searching permutation-based indexes for high dimensional data using inverted files. In this paper, three strategies for parallelization are discussed; posting lists decomposition, reference points decomposition, and multiple independent inverted files. We study performance, efficiency, and effectiveness of our strategies on high dimensional datasets of millions of images. Experimental results show a good performance compared to the sequential version with the same efficiency and effectiveness. © 2012 Springer-Verlag.