Batch Processing of Multi-Variant AES Cipher with GPU

Symmetric ciphers are key components in ensuring data confidentiality. One of the most well-known symmetric ciphers is AES cipher. Current research on accelerating AES cipher with GPU only focuses on execution of one variant of the cipher. To cater for multi-variant AES cipher execution, variant specific buffering needs to be implemented, which will pose latency overhead. In this paper, we present our design and implementation for batch execution of multi-variant AES cipher processing with GPU. The evaluation results show that GPU brings significantly higher throughput over the CPU-only implementation while maintaining result correctness. For data size of 512MB, we can achieve 3 times and 1.6 times higher performance over single core CPU and multi-core CPU, respectively.