Cryptographic algorithm acceleration using CUDA enabled GPUs in typical system configurations

The need to encrypt data is becoming more and more necessary. As the size of datasets continues to grow, the speed of encryption must increase to keep up or it will become a bottleneck. CUDA GPUs have been shown to offer performance improvements versus conventional CPUs for some data-intensive problems. This thesis evaluates the applicability of CUDA GPUs in accelerating the execution of cryptographic algorithms, which are increasingly used for growing amounts of data and thus will require significantly faster encryption and hashing throughput. Specifically, the CUDA environment was used to implement and experiment with three distinct cryptographic algorithms -- AES, SHA-2, and Keccak -- in order to show the applicability for various cryptographic algorithm classes. They were implemented in a system that emulates the conditions present in a real world environment, and the effects of offloading these tasks from the CPU to the GPU were assessed. Speedups up to 2.6x relative to the CPU were seen for single-kernel AES, but SHA-2 and Keccak did not perform as well as on the GPU as on the CPU. Multi-kernel AES saw speedups over single-kernel AES up to 1.4x, 1.65x, and 1.8x for two, three, and four kernels, respectively. This translates to speedups between 3.6x and 4.7x over CPU implementations of AES. Introducing a CPU load had a minimal effect on throughput whereas a GPU load was seen to decrease throughput by as much as 4%. Overall, CUDA GPUs appear to have potential for improving encryption throughputs if a parallelizable algorithm is selected