Performance Analysis of AES on CPU-GPU Heterogeneous Systems

AES is one of the most widely used encryption algorithms. With the ever-increasing amount of sensitive data that need protection, it is natural to turn to parallel AES solutions that exploit emerging architectures and reduce encryption time. In this paper, we present a collaborative implementation of AES (PAES-CPU-MultiGPU) for CPU-GPU heterogeneous systems. PAES-CPU-MultiGPU takes advantage of the hardware support for AES provided by modern CPU cores and, at the same time, benefits from the GPUs available on these systems. We compare the performance of our proposal with that of two other solutions that use only the CPU cores (PAES-CPU) and only the GPUs of the system (PAES-MultiGPU). The results reveal that PAES-CPU-MultiGPU achieves an overall performance similar to that of PAES-CPU, but using fewer CPU cores, outperforming by far PAES-MultiGPU. Also, PAES-CPU-MultiGPU outperforms PAES-CPU when an amount of CPU cores similar to that of commodity multicore machines is used.