High-performance Cryptographically Secure Pseudo-random Number Generation via Bitslicing
share
In this paper, a high-throughput Cryptographically Secure Pseudo-Random Number Generator (CSPRNG) using the bitslicing technique is proposed. In such technique, instead of the conventional row-major data representation, column-major data representation is employed which allows the bitsliced implementation to take full advantage of all the available datapath of the hardware platform. We use LFSR-based (Linear Feedback Shift Register) PRNG for our implementation since its register oriented architecture perfectly suits the GPU's many-core structure and allows for usage of bitslicing technique which can further improve its performance. In our GPU implementation, each GPU thread is capable of generating a remarkable number of 32 pseudo-random bits in each LFSR clock cycle. In order to obtain cryptographically suitable properties, we propose an SIMD vectorized fully parallel bitsliced implementation of the LFSR-based, cryptographically secure MICKEY 2.0 stream cipher algorithm for CSPRNG. To the best of our knowledge, our method not only achieves better performance, but also significantly outperforms optical solutions in terms of performance per cost while maintaining an acceptable measure of randomness. It should be mentioned that our implementation successfully passes the NIST test for statistical randomness and bitwise correlation criteria. Our proposed methodology significantly outperforms the current best implementations in the literature for computer-based PRNG. Moreover, our evaluations show 6.6x improvement over the Nvidia's proprietary high-performance PRNG, cuRAND library, achieving 5.2 Tb/s of throughput on the affordable Nvidia GTX 980 Ti.
READ FULL TEXT
