Evaluation and Analysis

The following is our statistics of experimentation. We got the data by testing the CPU time in OpenSSL and implementing DES and 3-DES on graphic card. Besides DES and 3-DES cryptography, AES cryptography had been implemented on CUDA in 2007 by Svetlin A. Manavski [28]. We will discuss how many differences between our results.

cipher \ compute method OpenSSL on CPU Implement DES on CUDA in our work

DES (64bits)

4M 60ms 32ms

8M 108ms 41ms

109.8M 2s 100ms 712ms

697M 14s 313ms 4s808ms

Table 4-7 Comparison DES in CPU and in GPU

35

Fig. 4-7 DES performance comparison

There are two parts in table 4-7. One is about the CPU time to implement DES in CPU and the other one is about the CPU time to implement DES in CUDA programming. We test the DES’s original CPU time in OpenSSL. It is obviously that we get better performance in CUDA programming. In plaintext of 4M byte, DES in CUDA programming has about 1/2 CPU time to DES of OpenSSL in CPU. In plaintext of 697M, we even get about three to four times of performance in CUDA programming. This means we would get more benefits in larger plaintext size to fit parallel algorithm.

36

Table 4-8 Comparison 3-DES in CPU and in GPU

Fig. 4-8 3-DES performance comparison

In 3-DES, we can see that it would get about five times performance in GPU than in CPU while the file size is near to 4M bytes. Moreover, it would get more than six times performance if the file size is larger than 700M bytes. The new 3-DES on

220 244

37

GPU is faster than AES on CPU in all the situations. This means that we can do the same encryption and decryption in 3-DES to get the similar performance. The following is relationship between AES on CPU and 3-DES on GPU.

In the related work about speeding up AES with CUDA programming by Svetlin A. Manavski, a comparison has been made between GPUs and a CPU implementation based on the OpenSSL [29] library. Both the internal GPU elaboration time is shown and the total time, that includes the time needed for the download and read-back operations (that means copying data from the host memory to the GPU device and back). The CUDA-AES [30] implementation on NVIDIA card is faster than the CPU on every input size with reference both to the internal GPU and to the total time [31].

A peak throughput rate of 8.28 Gbit/s is achieved with an input size of 8MB. In that case the GPU is 19.60 times faster than the CPU [32]. So we have both have about six times in GPU than in CPU when the file size is more than 512M bytes.

38

Chapter 5 Conclusion and Future works

5.1. Conclusion

Parallel programming nowadays becomes more and more important to programmer. This is because multi-cores is the trend to both CPU and GPU architecture development. However not all the programs are parallel program and this paper provide a method to solve this program. Find out the main points to parallelize a program is the most important thing to programmer.

CUDA code is not easy to optimize if the user do not understand the architecture of CUDA. Sometimes this makes programmer to abuse the graphic memory. Then the new CUDA program may have worse performance than original program on CPU. So the final performance is decided by programmer’s effort.

Although DES and 3-DES is an old cipher, there still many people use them to do encoding and decoding. This is because it needs a long time to verify the security of a cipher. DES and 3-DES however both have the trusty security. 3-DES is a compute-sensitive program, so it usually is used for off-time computing. For example, many finance institutions and banks use 3-DES to protect their system while making an inventory. Each cipher has its advantages and disadvantages, what we have to do is to properly use the advantages and keep of the disadvantages. It is better to ameliorate a defective cipher rather than abandon it rashly.

In the result we can see that CUDA programming is a cheap way to speed up a compute-sensitive program. Rather than implement a system on chip or on FPGA for about more than one hundred thousand, use just need to pay about five to eight

39

thousand to get CUDA. Maybe a program gets better performance on FPGA than on graphic card. But this is a cheap and easy method for a program to learn how to speed up a program or a system in parallel program.

5.2. Future work

(1) Speed up more cipher in CUDA programming

Except for DES and 3-DES, there are still many ciphers that we can try to speed them up in CUDA programming. But some cipher is not easy to implement on graphic card. For example the RSA cipher is very hard to get better performance in CUDA programming. RSA is an algorithm for public-key encryption. The encryption fun of

RSA is . C means cipher text, m and n are integers that 0 < m

< n. It is obviously that is very hard to parallelize the encryption function of RSA.

Although there is still some methods like squaring algorithm can improve this situation. However the performance is still too bad compared with programming in CPU and CUDA unless those programmers can design a new parallel algorithm to RSA. Entirely more high performance ciphers bring us more convenient life. It is necessary that programmers try to speed up more cipher in CUDA programming.

(2) Add the new 3-DES architecture to SSL

Secure Sockets Layer (SSL), are cryptographic protocols that provide security and data integrity for communications over networks such as the Internet. SSL encrypt the segments of network connections at the transport layer end-to-end. SSL contains three parts of cipher system, public-key cryptography, symmetric -secret system, and one-way hash function. 3-DES is one kind of public-key cryptography. If programmers furthermore implement the new 3-DES architecture to SSL, it may improve the all the transport layer performance.

40

(3) Improve the new 3-DES-website’s security

Our new 3-DES-website only offers data encryption and decryption. It could be safer if adding more safety mechanisms. However website security design is not our professional specialty. After this drawback has been reformed, the new 3-DES-websitr can be used to ensure internal management of enterprises.

41

Bibliography

[1] E.Biham and A. Shamir, “ Differential Cryptanalysis of DES-like Cryptosystems.” Journal of Cryptology, Vol.4, pp.3-72, (1911).

[2] National Institute of Standards and Technology (NIST), “FIPS 197: Advanced Encryption Standard (AES)”, 2001.

[3] E.Biham and A.Shamir, “ Differential Cryptanalysis of FEAL and N-Hash,

“ Advance in Cryptology – EUROCRYPT’91, Lecture Note in Computer Science, Vol.547, p.1-16, (1991).

[4] E.Biham and A.Shamir, “Differential Cryptanalysis of the full 16-round DES,

“ CRYPTO’92 Extended Abstracts, p.12-1-12-5, (1992).

[5] General Purpose Computation Using Graphics Hardware, http://www.gpgpu.org.

[6] NVidia CUDA , http://developer.NVidia.com/object/CUDA.html.

[7] OpenGL Shading Language Specification, Version 1.20

http://www.opengl.org/registry/doc/GLSLangSpec.Full.1.20.8.pdf

[8] AMD CTM, http://ati.amd.com/companyinfo/researcher/documents.html.

[9] NVIDIA CUDA Programming Guide, Version 0.8.2

http://developer.download.nvidia.com/compute/cuda/0_81/NVIDIA_CUDA_Pr ogramming_Guide_0.8.2.pdf

42

[10] I. Buck, A. Lefohn, P. McCormick, J. Owens, T. Purcell, R. Strodka, “General Purpose Computation on Graphics Hardware”. IEEE Visualization 05,

Minneapolis, USA, 2005.

[11] OpenGL Architecture Review Board, M. Woo, J. Neider, T. Davis, D. Shreiner,

“The OpenGL Programming Guide: The Official Guide to Learning OpenGL, Version 2”, 5th edition. ISBN 0321335732, Addison-Wesley, New York, 2005.

[12] D. L. Cook, A. D. Keroymytis, “Cryptographics: Exploiting Graphics Cards for Security”, Advancements in Information Security series, Springer, 2006.

[13] Svetlin A. Manavski, CUDA Compatible GPU as an Efficient Hardware Accelerator for AES Cryprography, Springer-Verlag, New York, 2007.

[14] A.Tardy-Corfdir and H.Gilbert, “A Known Plaintext Attack of FEAL-4 and FEAL-6, “ Advances in Cryptology – CRYPTO’91, Lecture Notes in Computer Science, Vol.576, (1991).

[15] M. Matsui and A. Yamagishi, “A New Method for Known Plaintext Attack of FEAL Cipher, “ Advances in Cryptology – EUROCRY’92, Lecture Notes in Computer Science, Vol.658 , (1992).

[16] A. Shamir, “ On the Security of DES, “ Advances in Cryptology –CRYPTO’85, Lecture Notes in Computer Science, (1985).

[17] R.A.Rueppel, “Analysis and Design of stream Cipher, “Springer Verlag, (1986).

43

[18] Federal Information Processing Standards Publication (FIPS PUB) 46-2, Data Encryption Standard (DES), National Institute of Standards and Technology, Washington, DC, 1993.

[19] M. J. Wiener, “Efficient DES Key Search,” Technical Federal Information Processing Standards Publication (FIPS PUB) 46-2, Data Encryption Standard (DES), National Institute of Standards and Technology, Washington, DC, 1993.

[20] M. J. Wiener, “Efficient DES Key Search,” Technical Report TR-244, School of Computer Science, Carleton University, Ottawa, Canada, May 1994.

Presented at the rump session of Crypto’93.

[21] F. Hendessi and M. R. Aref, “A Successful Attack Against the DES,”

Information Theory and Application, Third Canadian Workshop Proceedings,

1994, pp. 78-90.

[22] Frank Rubin, “Foiling an Exhaustive Key-Search Attack,” CRYPTOLOGIA 11, No. 2, 102-107 (April 1987).

[23] M. Hellman, “A Cryptanalytic Time-Memory Trade-off,” IEEE Trans. Info.

Theory IT-26, No. 4, 401-406 (1980).

[22] E. Biham and A. Shamir, Differential Cryptanalysis of the Data Encryption Standard, Springer-Verlag, New York, 1993.

[24] E. Biham and A. Shamir, Differential Cryptanalysis of the Data Encryption Standard, Springer-Verlag, New York, 1993.

44

[25] M. Matsui, “The First Experimental Cryptanalysis of the Data Encryption Standard,” Lecture Notes in Computer Science 839, Advances in

Cryptology-Crypto ’94 Proceedings, Springer-Verlag, New York, 1994.

[26] D. Coppersmith, “The Data Encryption Standard (DES) and Its Strength Against Attacks,” IBM J. Res. Devebp. 38, NO. 3, 243-250 (1994).

[27] B. S. Kaliski, Jr. and M. J. B. Robshaw, “Multiple Encryption: Weighing Security and Performance,” Dr. Dobb’s Journal 21, No. 1, 123-127 (1996)

[28] J. Daemen, V. Rijmen, “AES Proposal: Rijndael”. Original AES Submission to NIST, 1999.

[29] OpenSSL Open Source Project, http://www.openssl.org.

[30] C. Su, T. Lin, C. Huang, C. Wu, “A High-Throughput Low-Cost AES

processor”. IEEE Communications Magazine, vol. 41, no. 12, pp. 86-91, 2003.

[31] J. Wolkerstorfer, E. Oswald, M. Lamberger, “An ASIC Implementation of the AES Sboxes”. RSA Conference 02, San Jose CA, 2002.

[32] A. Hodjat, I. Verbauwhede, “Minimum Area Cost for a 30 to 70 Gbits/s AES Processor”. IEEE Computer Society Annual Symposium on VLSI (ISVLSI 2004), Eerging Trends in VLSI System Design, pp 83-88, 2004.