AES-128 ECB encryption on GPUs and effects of input plaintext patterns on performance

Abstract

In the recent years, the Graphics Processing Units (GPUs) have gained popularity for general purpose applications, immensely outperforming traditional optimized CPU based implementations. A class of such applications implemented on GPUs to achieve faster execution than CPUs include cryptographic techniques like the Advanced Encryption Standard (AES) which is a widely deployed symmetric encryption/decryption scheme in various electronic communication domains. With the drastic advancements in electronic communication technology, and growth in the user space, the size of data exchanged electronically has increased substantially. So, such cryptographic techniques become a bottleneck to fast transfers of information. In this work, we implement the AES-128 ECB Encryption on two of the recent and advanced GPUs (NVIDIA Quadro FX 7000 and Tesla K20c) with different memory usage schemes and varying input plaintext sizes and patterns. We obtained a speedup of up to 87x against an advanced CPU (Intel Xeon X5690) based implementation. Moreover, our experiments reveal that the different degrees of pattern repetitions in input plaintext affect the encryption performance on GPU.