GPU exhibits the capability for applications with a high level of parallelism despite its low cost. The support of integer and logical instructions by the latest generation of GPUs enables us to implement cipher algorithms more easily. However, decisions such as parallel processing granularity and memory allocation impose a heavy burden on programmers. Therefore, this paper presents results of several experiments that were conducted to elucidate the relation between memory allocation styles of variables of AES and granularity as the parallelism exploited from AES encoding processes using CUDA with an NVIDIA GeForce GTX285 (Nvidia Corp.). Results of these experiments showed that the 16 bytes/thread granularity had the highest performance. It achieved approximately 35 Gbps throughput. It also exhibited differences of memory allocation and granularity effects around 2%-30% for performance in standard implementation. It shows that the decision of granularity and memory allocation is the most important factor for effective processing in AES encryption on GPU. Moreover, implementation with overlapping between processing and data transfer yielded 22.5 Gbps throughput including the data transfer time.

No abstract

As the data protection with encryption becomes important day by day, the encryption processing using General Purpose computation on a Graphic Processing Unit (GPGPU) has been noticed as one of the methods to realize high-speed data protection technology. GPUs have evolved in recent years into powerful parallel computing devices, with a high cost-performance ratio. However, many factors affect GPU performance. In earlier work to gain higher AES performance using GPGPU in various ways, we obtained the following two technical viewpoints: (1) 16 Bytes/Thread is the best granularity (2) Extended key and substitution table stored in shared memory and plaintext stored in register are the best memory allocation style.However, AES is not the only cipher algorithm widely used in the real world. For this reason, this study was undertaken to test the hypothesis that these two findings are applicable to implementation of other symmetric block ciphers on two generation of GPU. In this study, we targeted five 128-bit symmetric block ciphers, AES, Camellia, CIPHERUNICORN-A, Hierocrypt-3, and SC2000, from an e-government recommended ciphers list by the CRYPTography Research and Evaluation Committees (CRYPTREC) in Japan. We evaluated the performance of these five symmetric block ciphers on the machine including a 4-core CPU and each GPU using three method: (A) throughput without data transfer, (B) throughput with data transfer and overlapping encryption processing on GPU, (C) throughput with data transfer and non-overlapping encryption processing on GPU. Results demonstrate that the throughput of implementation of SC2000 in method (A) on Tesla C2050 achieved extremely high 73.4 Gbps. Additionally, the throughput obtained using methods (B) and (C) deteriorated to 33.4 Gbps and 18.3 Gbps, respectively. Method (B) showed effective throughput with an approximately 4.7 times higher speed compared to that obtained when using 8 threads on a 4-core CPU.

No abstract

No abstract