Parallel Computing Mode in Homomorphic Encryption Using GPUs Acceleration in Cloud

Abstract

As an important encryption algorithm of performing computations directly on ciphertext without any need of decryption and compromising privacy, homomorphic encryption is an increasingly popular research topic of protecting privacy of the data in cloud security research. However, it will be a heavy workload for resources as the high computational complexity of homomorphic encryption. Therefore, GPU acceleration is employed to speed up homomorphic encryption. Motivated by this observation, we utilize parallel computing mode in DGHV algorithm with GPUs acceleration based on CPU-GPUs hybrid system. Our main contribution is to present a parallel computing mode for large-scale data encryption based on CPU-GPUs hybrid system as fast as possible. Specifically, we applies parallel computing mode in DGHV with GPUs acceleration to reduce the time duration and provide a comparative performance study. We further design pipeline architecture of processing stream to accelerate the speed of DGHV algorithm. Furthermore, experimental results validate that different parallelism has the corresponding granularity in parallel computing mode. Experimental results show that our method gains more than 84% improvement (run time), 67% improvement (run time), and 80% improvement (run time) compared to the sequential data encryption, sequential homomorphic addition, and sequential homomorphic multiplication in DGHV algorithm respectively.