Efficient Cryptographic Processing Using On-Chip GPUs in System-on-Chip Architectures

Efficient Cryptographic Processing Using On-Chip GPUs in System-on-Chip Architectures

연구 분야: Cryptography

학회: 2023 Eleventh International Symposium on Computing and Networking Workshops (CANDARW)

In recent years, with the spread of telework, there has been an increase in demand for encryption when working on laptops at home or in the office. However, since the performance is lower than that of desktop PCs, the CPU load increases, thereby reducing overall performance. Therefore, we propose a system that offloads encryption processing to the GPU in the SoC, so that it can be done even on notebook PCs. As a result of evaluation experiments, we found that the GPU kernel implemented in this research is faster than the CPU implementation if the size of the data to be encrypted is above a certain level, and has a maximum throughput of about 10.3 times higher, so that sufficient effects can be obtained. Therefore, we can expect to improve processing efficiency by devising scheduling, such as using different CPUs and GPUs depending on the data size. However, when compared to cryptographic processing using AES-NI, it was confirmed that the throughput was approximately 13.3 times lower even at the point where the GPU implementation showed the highest Gbps. We also confirmed that this system can perform encryption without occupying the CPU.