A Controllable Pipeline Framework of Block Ciphers on GPU for Streaming Data

Abstract

With the application and development of real-time multimedia information technology such as online meetings and short videos, the demand for secure transmission of streaming media data based on high throughput and low latency has increased significantly. Since servers must encrypt large amounts of data in real-time without compromising their service capabilities, Graphic Processing Units (GPUs) are considered the best candidates for cryptographic accelerators to handle large amounts of data in this situation. For the characteristics of streaming data, this paper proposes a GPU-based controllable pipelined encryption framework, which guarantees security during streaming data transmission while satisfying the requirements of high throughput and low latency. Firstly, in order to make the CPU in the traditional heterogeneous system understand the intermediate state of the GPU’s operation in real-time, this paper adopts the cyclic checkpoint mechanism to realize the interrupt control; secondly, real-time signal feedback between the host and the device each other during the execution of the kernel function is realized by real-time monitoring and other modules; through the above ways, the CPU-GPU pipeline control framework is realized. Finally, based on AES and SM4 encryption algorithms, this paper designs and realizes the pipeline encryption and decryption process of streaming data based on CUDA in Python environment, and verifies the group automation pipeline framework based on interrupt monitoring and controlling through experiments; and under the condition of using 5 threads grouping, the data throughput rate reaches 1.01Gbps, which has a high throughput rate.

Keywords

• Graphic processing units (GPUs)
• pipeline computing
• block cipher
• CUDA