Evaluating Encryption Overhead and Offload Techniques in High-Speed Data Transfers

The increasing demand for secure, high-speed data transfers in scientific and data-intensive applications presents a major challenge due to the significant processing overhead introduced by encryption. This paper evaluates the performance impact of encryption on GridFTP data transfers over a 100 Gbps link, focusing on throughput degradation, CPU utilization, and fairness in both encrypted and non-encrypted traffic scenarios. The experiments presented in this paper examine the influence of propagation delays and TCP Congestion Control Algorithms (CCAs), revealing that software-based encryption imposes severe CPU bottlenecks that limit scalability and link utilization. To address these limitations, this paper explores acceleration techniques that include CPU-based methods, such as Intel AES-NI, and hardware-assisted solutions, such as SmartNIC offloading. Results show that SmartNICs maximize throughput, achieving higher scalability compared to traditional approaches.