DESIGN AND ACCELERATION OF FULLY HOMOMORPHIC ENCRYPTION IN THE RNS- CKKS SCHEME USING GPU PARALLELIZATION

• Main Author: Jedidia Augustine, Jevant
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81792
• Institution: Institut Teknologi Bandung

Homomorphic encryption is an encryption scheme that allows computation to be performed on ciphertext. One of the most widely used homomorphic encryption schemes is CKKS. The ciphertext of the CKKS scheme is representated in polynomial form. For good security, the degree and coefficients of said polynomial need to have high values. For this reason, CKKS was further developed into RNS-CKKS in order to increase the efficiency of operations and data storage. However, some operations, such as ciphertext multiplication, still require high execution time, especially for higher polynomial degree. In order to solve this issue, parallelization of the RNS-CKKS scheme is used to accelerate the process. The serial implementation of the scheme is done using C++, while the parallel implementation is done using CUDA. Two main techniques used to optimize the parallel implementation are array flattening and kernel fusion. Testing is done by comparing the execution time of serial and parallel implementation and then calculating the speedup or slowdown of each operation. As a result, the NTT, iNTT, encryption, decryption, and multiplication processes experienced accelerations of 3.7x, 3.7x, 3.26x, 3.96x, and 4.68x, respectively. Meanwhile, the decoding and addition processes experienced a slowdowns of 0.57x dan 0.86x, respectively. No significant acceleration was observed during the encoding process.