Title: The GPU as an indirection engine for a fast information retrieval

Authors: Manuel Ujaldon, Joel Saltz

Addresses: Computer Architecture Department, University of Malaga, Complejo Tecnologico, Campus Teatinos, Malaga 29071, Spain, ' Biomedical Informatics Department, Ohio State University, 3197 Graves Hall, 333 W, 10th Ave, Columbus, OH 43210, USA

Abstract: E-commerce database and web servers are distinguished for accessing large indexed data sets. The paper proposes a novel approach for exploiting the graphical resources of a commodity PC, transforming the graphics pipeline (GPU) into an indirection engine for a fast retrieval of indexed data. We show how a GeForce FX graphics card can make use of vertices, textures and colours to solve up to six nested indexed lookups entirely in hardware, achieving a performance gain of up to 400% vs. a Pentium 4 processor with five times higher clock frequency. As they are evolving, GPUs are converting into general-purpose processors willing to cooperate with the CPU for a fast joint execution. Our results demonstrate the efficiency of the graphics processor for accessing indexed information, suggesting its extensive use in web servers, relational databases and XML systems, where multiple queries can run simultaneously on both processors.

Keywords: web servers; database systems; information retrieval; indirection engine; indexed data accessing; streaming computation; graphics processing; OpenGL; e-commerce; e-business; electronic commerce; electronic business; GPUs.

DOI: 10.1504/IJEB.2005.007275

International Journal of Electronic Business, 2005 Vol.3 No.3/4, pp.316 - 327

Published online: 30 Jun 2005 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article