Skip to main content
SpringerLink
Log in

Combining CPU and GPU architectures for fast similarity search

  • Published:
Distributed and Parallel Databases Aims and scope Submit manuscript

Abstract

The Signature Quadratic Form Distance on feature signatures represents a flexible distance-based similarity model for effective content-based multimedia retrieval. Although metric indexing approaches are able to speed up query processing by two orders of magnitude, their applicability to large-scale multimedia databases containing billions of images is still a challenging issue. In this paper, we propose a parallel approach that balances the utilization of CPU and many-core GPUs for efficient similarity search with the Signature Quadratic Form Distance. In particular, we show how to process multiple distance computations and other parts of the search procedure in parallel, achieving maximal performance of the combined CPU/GPU system. The experimental evaluation demonstrates that our approach implemented on a common workstation with 2 GPU cards outperforms traditional parallel implementation on a high-end 48-core NUMA server in terms of efficiency almost by an order of magnitude. If we consider also the price of the high-end server that is ten times higher than that of the GPU workstation then, based on price/performance ratio, the GPU-based similarity search beats the CPU-based solution by almost two orders of magnitude. Although proposed for the SQFD, our approach of fast GPU-based similarity search is applicable for any distance function that is efficiently parallelizable in the SIMT execution model.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9
Fig. 10
Fig. 11
Fig. 12
Fig. 13
Fig. 14
Fig. 15
Fig. 16
Fig. 17
Fig. 18

Similar content being viewed by others

Notes

  1. In theory, two subsequent blocks dispatched to the same GPU device may overlap in some operations. Modern GPUs have independent units for host-device memory transfers, therefore it should be possible to overlap data transfer and SQFD computation of two subsequent blocks. In order to do so, the size of the block needs to be restricted so that at least two data blocks would fit the GPU device memory. Unfortunately, we have encountered many technical problems when attempting to pipeline execution and data transfers. It is our belief that these problems are caused by flaws in hardware drivers and/or OpenCL implementation.

  2. We use a kind of schema together with a conceptual explanation of the algorithm, because a code listing in parallel framework would be not as concise and easy to read.

  3. As mentioned before, the larger the better.

  4. Actually, these constants are suitable only for α=0.2 and α=0.01. The value α=3 requires the largest possible blocks since it does not benefit much from indexing.

  5. It is safe to say that modern GPUs have sufficient memory capacity to accommodate pivot tables for databases that fit the host memory of an ordinary server.

References

  1. Beecks, C., Lokoč, J., Seidl, T., Skopal, T.: Indexing the signature quadratic form distance for efficient content-based multimedia retrieval. In: Proc. ACM Int. Conf. on Multimedia Retrieval, pp. 24:1–24:8 (2011)

    Google Scholar 

  2. Beecks, C., Seidl, T.: On stability of adaptive similarity measures for content-based image retrieval. In: MMM, pp. 346–357 (2012)

    Google Scholar 

  3. Beecks, C., Uysal, M.S., Seidl, T.: Signature quadratic form distances for content-based similarity. In: Proc. ACM Multimedia, pp. 697–700 (2009)

    Google Scholar 

  4. Beecks, C., Uysal, M.S., Seidl, T.: A comparative study of similarity measures for content-based multimedia retrieval. In: Proc. IEEE ICME, pp. 1552–1557 (2010)

    Google Scholar 

  5. Beecks, C., Uysal, M.S., Seidl, T.: Signature quadratic form distance. In: Proc. ACM CIVR, pp. 438–445 (2010)

    Google Scholar 

  6. Bolettieri, P., Esuli, A., Falchi, F., Lucchese, C., Perego, R., Piccioli, T., Rabitti, F.: CoPhIR: a test collection for content-based image retrieval. 0905.4627v2 (2009). http://cophir.isti.cnr.it

  7. Bustos, B., Deussen, O., Hiller, S., Keim, D.: A graphics hardware accelerated algorithm for nearest neighbor search. In: Computational Science—ICCS 2006, pp. 196–199 (2006)

    Chapter  Google Scholar 

  8. Chávez, E., Navarro, G., Baeza-Yates, R., Marroquín, J.L.: Searching in metric spaces. ACM Comput. Surv. 33(3), 273–321 (2001). doi:10.1145/502807.502808

    Article  Google Scholar 

  9. Deselaers, T., Keysers, D., Ney, H.: Features for image retrieval: an experimental comparison. Inf. Retr. 11(2), 77–107 (2008). doi:10.1007/s10791-007-9039-3

    Article  Google Scholar 

  10. Garcia, V., Debreuve, E., Barlaud, M.: Fast k nearest neighbor search using gpu. In: IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, CVPRW’08, pp. 1–6. IEEE, New York (2008)

    Google Scholar 

  11. Geusebroek, J.M., Burghouts, G.J., Smeulders, A.W.M.: The Amsterdam library of object images. Int. J. Comput. Vis. 61(1), 103–112 (2005)

    Article  Google Scholar 

  12. Hafner, J., Sawhney, H.S., Equitz, W., Flickner, M., Niblack, W.: Efficient color histogram indexing for quadratic form distance functions. IEEE Trans. Pattern Anal. Mach. Intell. 17, 729–736 (1995). doi:10.1109/34.391417

    Article  Google Scholar 

  13. Hetland, M.L.: The basic principles of metric indexing. In: Coello, C.A.C., Dehuri, S., Ghosh, S. (eds.) Swarm Intelligence for Multi-objective Problems in Data Mining. Studies in Computational Intelligence, vol. 242. Springer, Berlin (2009)

    Chapter  Google Scholar 

  14. Hetland, M.L.: Ptolemaic indexing. arXiv:0911.4384 [cs.DS] (2009)

  15. Hu, R., Rüger, S., Song, D., Liu, H., Huang, Z.: Dissimilarity measures for content-based image retrieval. In: Proc. IEEE International Conference on Multimedia & Expo, pp. 1365–1368 (2008). doi:10.1109/ICME.2008.4607697

    Google Scholar 

  16. Huiskes, M.J., Lew, M.S.: The mir flickr retrieval evaluation. In: Proc. ACM MIR, pp. 39–43 (2008)

    Google Scholar 

  17. Leow, W.K., Li, R.: The analysis and applications of adaptive-binning color histograms. Comput. Vis. Image Underst. 94(1–3), 67–91 (2004). doi:10.1016/j.cviu.2003.10.010

    Article  Google Scholar 

  18. Lieberman, M., Sankaranarayanan, J., Samet, H.: A fast similarity join algorithm using graphics processing units. In: IEEE 24th International Conference on Data Engineering, ICDE 2008, pp. 1111–1120. IEEE, New York (2008)

    Chapter  Google Scholar 

  19. Lokoč, J., Hetland, M., Skopal, T., Beecks, C.: Ptolemaic indexing of the signature quadratic form distance. In: Proceedings of the Fourth International Conference on Similarity Search and Applications, pp. 9–16. ACM, New York (2011)

    Google Scholar 

  20. Lokoč, J.: Cloud of points generator. SIRET Research Group (2010). http://siret.ms.mff.cuni.cz/projects/pointgenerator/

  21. Mémoli, F., Sapiro, G.: Comparing point clouds. In: SGP’04: Proceedings of the 2004 Eurographics/ACM SIGGRAPH Symposium on Geometry Processing, pp. 32–40. ACM, New York (2004). doi:10.1145/1057432.1057436

    Chapter  Google Scholar 

  22. Mico, M.L., Oncina, J., Vidal, E.: A new version of the nearest-neighbour approximating and eliminating search algorithm (aesa) with linear preprocessing time and memory requirements. Pattern Recognit. Lett. 15(1), 9–17 (1994). doi:10.1016/0167-8655(94)90095-7

    Article  Google Scholar 

  23. Mikolajczyk, K., Schmid, C.: A performance evaluation of local descriptors. IEEE Trans. Pattern Anal. Mach. Intell. 27(10), 1615–1630 (2005). doi:10.1109/TPAMI.2005.188

    Article  Google Scholar 

  24. Navarro, G.: Analyzing metric space indexes: what for? In: IEEE SISAP 2009, pp. 3–10 (2009)

    Google Scholar 

  25. NVIDIA: Fermi GPU Architecture. http://www.nvidia.com/object/fermi_architecture.html

  26. Pan, J., Manocha, D.: Fast gpu-based locality sensitive hashing for k-nearest neighbor computation. In: Proceedings of the 19th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, pp. 211–220. ACM, New York (2011)

    Google Scholar 

  27. Puzicha, J., Buhmann, J., Rubner, Y., Tomasi, C.: Empirical evaluation of dissimilarity measures for color and texture. In: Proc. IEEE International Conference on Computer Vision, vol. 2, pp. 1165–1172 (1999)

    Chapter  Google Scholar 

  28. Rubner, Y., Tomasi, C., Guibas, L.J.: The earth mover’s distance as a metric for image retrieval. Int. J. Comput. Vis. 40(2), 99–121 (2000). doi:10.1023/A:1026543900054

    Article  MATH  Google Scholar 

  29. Samet, H.: Foundations of Multidimensional and Metric Data Structures. Morgan Kaufmann, San Mateo (2006)

    MATH  Google Scholar 

  30. Skopal, T., Bartoš, T., Lokoč, J.: On (not) indexing quadratic form distance by metric access methods. In: Proc. Extending Database Technology (EDBT). ACM, New York (2011)

    Google Scholar 

  31. Zezula, P., Amato, G., Dohnal, V., Batko, M.: Similarity Search: The Metric Space Approach. Advances in Database Systems. Springer, New York (2005)

    Google Scholar 

Download references

Acknowledgements

This research has been supported by Czech Science Foundation (GAČR) project 202/11/0968, by the grant agency of Charles University (GAUK) project no. 277911, and by the Deutsche Forschungsgemeinschaft within the Collaborative Research Center SFB 686.

Author information

Authors and Affiliations

  1. SIRET Research Group, Faculty of Mathematics and Physics, Charles University in Prague, Prague, Czech Republic

    Martin Kruliš, Tomáš Skopal & Jakub Lokoč

  2. Data Management and Data Exploration Group, RWTH Aachen University, Aachen, Germany

    Christian Beecks

Authors
  1. Martin Kruliš
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Tomáš Skopal
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Jakub Lokoč
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Christian Beecks
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Tomáš Skopal.

Additional information

Communicated by: Kaushik Chakrabarti.

Rights and permissions

Reprints and permissions

About this article

Cite this article

Kruliš, M., Skopal, T., Lokoč, J. et al. Combining CPU and GPU architectures for fast similarity search. Distrib Parallel Databases 30, 179–207 (2012). https://doi.org/10.1007/s10619-012-7092-4

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10619-012-7092-4

Keywords

Search

Navigation