AsyncGBP: Unleashing the Potential of Heterogeneous Computing for SSL/TLS with GPU-based Provider

The proliferation of IoT and 5G technologies has led to an explosion of data traffic that data centers must handle while ensuring secure transmission via SSL/TLS. The high volume of cryptographic operations required imposes performance bottlenecks. The GPU-based cryptographic accelerator is one of the competitive solutions. However, significant structural differences with practical applications confine their capacities to specific domains, such as offline cryptanalysis, undermining their potential for real-world cryptographic acceleration. This paper investigates the feasibility of using GPUs as cryptographic accelerators for concurrent data secure transmission scenarios like SSL/TLS. Specifically, we propose AsyncGBP, a framework that integrates the original OpenSSL software stack with the heterogeneous GPU-based accelerator. To enhance user-friendliness and take full advantage of GPUs' SIMT execution model, AsyncGBP features an OpenSSL-compatible asynchronous design, which seamlessly converts cryptographic requests from synchronous to asynchronous mode, efficiently aggregates numerous requests, and rationally schedules GPU for computation. We also provide a finegrained GPU-based cryptographic algorithm stack that includes X25519, Ed25519, and ChaCha20-Poly1305. A comprehensive evaluation shows that AsyncGBP can efficiently achieve up to 97% of GPU local performance on an RTX 3070, resulting in an improvement of up to 137x compared to the default OpenSSL provider in a single-process setting. Furthermore, AsyncGBP outperforms the existing fastest commercial-off-the-shelf OpenSSL-compatible TLS accelerator by a significant margin, achieving a 5.3x to 7.0x performance improvement.