Homomorphic Encryption Offloading And Its Application In Privacy-Preserving Computing

With the advent of the era of big data,fields such as artificial intelligence are in-creasingly relying on data to drive.All aspects of society and life are affected by data science represented by data mining.As an industrial raw material in the era of big data,data has become an important asset of contemporary society.In order to make data better play its value,the free exchange and sharing of data from all walks of life has be-come an urgent need in the era of big data.However,in recent years,security incidents have occurred frequently.The threats of data security and privacy leakage have be-come the main obstacles to data circulation.Privacy-preserving computing technology has become an important cornerstone for the further development of big data.As an important tool for privacy-preserving computing,homomorphic encryption can be directly calculated on high-security ciphertext to avoid the risk of privacy leak-age.However,due to high computational complexity,homomorphic encryption is cur-rently difficult to apply to large-scale computing.This thesis attempts to accelerate homomorphic encryption from the perspective of hardware offload,and provides the possibility for the application of homomorphic encryption in mass production.Intel QuickAssist Technology(QAT)is a hardware accelerator launched by In-tel for accelerating conventional encryption and compression operations.This thesis proposes an efficient asynchronous homomorphic encryption offloading framework QHCS based on QAT,offloading homomorphic encryption to QAT to achieve accel-erate.QHCS achieves efficient asynchronous offloading by reconstructing the ho-momorphic encryption software stack and introducing fibre technology.In addition,QHCS also provides two different schemes for different performance metrics prefer-ences(throughput,delay)of the application.Further,based on QHCS,this thesis com-pletely implements a privacy-preserving ridge regression application for data trading.Experiments show that based on QHCS,the homomorphic encryption algorithm Paillier is 110 times higher in performance than software.The privacy-preserving ridge regression on millions of high-dimensional data only takes more than ten minutes,which can meet the needs of large-scale privacy-preserving computing.