| Presentation | 2010-01-27 Granularity Optimization Method for AES Encryption Implementation on CUDA Naoki NISHIKAWA, Keisuke IWAI, Takakazu KUROKAWA, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | GPGPU as parallel computation platform has been noticed from almost all reseach fields. In particular CUDA occupies a high share of the GPGPU development environment. With CUDA, programmers are responsible for deciding the number of threads or thread blocks, but the optimum value is actually obtained by programmers' repetitive experiment. As the result, we have attempted to construct an automatic optimization model based on the number of threads used. As a first step, this paper presents analysis of how combinations such as data type of plaintext, memory allocation style of plaintext, and granularity, affect GPU performance. These experimental results show that there is up to a 6.6-fold performance increase among implementation methods with such combinations, resulting in the following insights: (1) Securing the number of threads, before the implementation of memory access optimization, is necessary. (2) 16Byte/Thread granularity leads to higher GPU performance than 4Byte/Thread and 1Byte/Thread granularity. (3) Different data types in plaintext, memory allocation styles of plaintext, and granularity affect GPU performance. In addition, we confirmed AES encryption method with 4Byte/Thread granularity for plaintexts, stored in shared memory as both unsigned integer and structure of array leads to the GPU's maximum performance and this implementation method achieved as apploximately 47-fold speed up as normal AES implemetation on Core i7-920 2.66GHz CPU. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | GPGPU / CUDA / Cipher / AES / Performance predication |
| Paper # | VLD2009-86,CPSY2009-68,RECONF2009-71 |
| Date of Issue |
| Conference Information | |
| Committee | CPSY |
|---|---|
| Conference Date | 2010/1/19(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Computer Systems (CPSY) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Granularity Optimization Method for AES Encryption Implementation on CUDA |
| Sub Title (in English) | |
| Keyword(1) | GPGPU |
| Keyword(2) | CUDA |
| Keyword(3) | Cipher |
| Keyword(4) | AES |
| Keyword(5) | Performance predication |
| 1st Author's Name | Naoki NISHIKAWA |
| 1st Author's Affiliation | Department of Computer Science, National Defense Academy() |
| 2nd Author's Name | Keisuke IWAI |
| 2nd Author's Affiliation | Department of Computer Science, National Defense Academy |
| 3rd Author's Name | Takakazu KUROKAWA |
| 3rd Author's Affiliation | Department of Computer Science, National Defense Academy |
| Date | 2010-01-27 |
| Paper # | VLD2009-86,CPSY2009-68,RECONF2009-71 |
| Volume (vol) | vol.109 |
| Number (no) | 394 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |