A Comparison of CPU and OpenCL Parallelization Methods for Correlation and Graph Layout Algorithms used in the Network Analysis of High Dimensional Data
The Roslin Institute and R(D)SVS, University of Edinburgh, Easter Bush, Midlothian, EH25 9RG, Scotland, UK
University of Edinburgh, 2011
@article{theocharidis2011comparison,
title={A Comparison of CPU and OpenCL Parallelization Methods for Correlation and Graph Layout Algorithms used in the Network Analysis of High Dimensional Data},
author={Theocharidis, A. and Hemani, G. and Kargas, M. and Freeman, T.C.},
year={2011}
}
MOTIVATION: Many algorithms used in analysis of high dimensional data require significant processing time due to the sheer number of values compared. We describe the results of the parallelization of two algorithms central to the functionality of the network analysis tool BioLayout Express 3D; the calculation of correlation (Pearson, Spearman Rank) coefficient matrices used to define relationships in large datasets, such as between gene expression profiles in microarray analyses and the Fruchterman-Rheingold graph layout algorithm used in the visualization of the resulting networks. RESULTS: Initially, the Java 1.6 and ANSI C99 languages were used to provide multithreaded implementations of these algorithms and to run on all available CPUs. Secondly, the OpenCL C language was used as part of the OpenCL API to harness the processing power of GPUs. Both approaches have been implemented using a platform and hardware independent approach. We discuss the issues associated with the parallelization of these very different algorithms and provide detailed comparisons of the results where we have achieved speed-ups of more than 60x times compared to non-parallel implementations. AVAILABILITY: The code is publicly available and utilized within the current release (version 2.0) of BioLayout Express 3D (www.biolayout.org).
February 5, 2013 by hgpu
