![](bg6.png)
Table 4: Computing duration using nVidia CUDA enabled
video cards.
Video Card Capabilities Duration (ms)
nVidia Quadro
NV 135 M
2 Multiprocessors 23625
nVidia
GeForce 9500
GT
4 Multiprocessors 7266
nVidia
GeForce
GTX275
30 Multiprocessors 1390
An on-line authentication system, using several
graphics cards to process distances between
invariants strings, will have the architecture
presented below.
Figure 9: Authentication system using multiple nVidia
CUDA enabled video cards.
The communication through the PCI Express
BUS is transparent to the developer because it is
realised by the provided CUDA driver (nVidia
2008). The implementing problems are similar to
ones that appear when implementing the server
solution. The main advantage of this solution is that
the video cards are relatively cheap and their
applicability area is rather large. The system is also
scalable, because adding extra graphic cards is
relatively easy, without major code modifications.
4 CONCLUSIONS
We have presented 3 solutions of accelerating on-
line authentication, by using dynamic handwritten
signature. We have presented the signature
processing which is made, and we have shown 3
methods of speeding-up the most computational
blocks. The following table synthesizes the results
we have obtained.
We have considered a system using a 16
processor powerful server, one using 3 USB 2.0
FPGA boards with 10 Levenshtein processors each,
working at 300 MHz and also a system using a
single nVidia GeForce GTX275 video card.
Table 5: Comparison between the proposed acceleration
solutions.
Solution Price Duration (ms)
16 Processor Server ~8000 USD 1328
3 FPGA Boards on a PC ~2000 USD 1400
1 nVidia GeForce GTX275
video card
~300 USD 1390
As we can see from the table above, the solution
that should be used is obvious. Motherboards built
using nVidia SLI technology allow up to 3 video
cards on one single system so the speed achieved
can be highly improved with minimal costs.
Given this context, an on-line system of
authenticating users by their dynamic signature, can
respond to a number of around 100 requests per
second, when using 3 video cards, which makes it a
high security feature needed to be considered. Using
the proposed architectures, the system is very
scalable and if the number of requests increases,
more computing power can be added at a small
price.
REFERENCES
Marcu, E., 2009. Method of combining the degrees of
similarity in handwritten signature authentication,
using neural networks. In AI-2009, The Twenty-ninth
SGAI International Conference Cambridge, UK.
Springer
Marcu, E., 2009. Self-built grid. In IDC’2009, 3rd
International Symposium on Intelligent Distributed
Computing. Springer
Hoang, D. T., Lopresti, D., 1993. FPGA Implementation
of Systolic Sequence Alignment. In International
Workshop on Field Programmable Logic and
Applications. Springer Berlin
Mohd, E. T., Mohd, Y. I. I., Tee., H. H., Madhihah, S.,
2008. Hardware based SPAM/UCE Filter Design with
Levenshtein Distance Algorithm: A Framework. In
Proceedings of Internet Convergence Conference, 11-
13 March 2008, Kuala Lumpur. Non-Scopus Cited
Publication.
Manavski, A. S., Valle. G., 2008. CUDA compatible GPU
cards as efficient hardware accelerators for Smith-
Waterman sequence alignment. In BMC
Bioinformatics 2008. BMC Bioinformatics.
nVidia CUDA Zone Examples of GPU Processing,
http://www.nvidia.co.uk/object/cuda_home_uk.html
nVidia CUDA Programming Guide, 2008. http://
developer.download.nvidia.com/compute/cuda/2_3/to
olkit/docs/NVIDIA_CUDA_Programming_Guide_2.3
.pdf
ICEIS 2010 - 12th International Conference on Enterprise Information Systems
126