The Key Technology Of Performance Optimization Of Zero Knowledge Proof For Heterogeneous Architecture

Zero-knowledge proofs allow the prover to prove the correctness of a statement to a verifier without giving away any other information.As a cryptographic protocol capable of encrypting and compressing information,zero-knowledge proof has been used in recent years in the blockchain field to solve privacy protection and throughput problems.At present,the performance of zero-knowledge proof is still an important problem hindering its application.The existing works mainly use custom accelerators to improve the performance of zero-knowledge proof,but there is a lack of research on the performance of zero-knowledge proof for the most widely used CPU-GPU architecture.In this paper,we establish a performance analysis model for the zero-knowledge proof on CPU-GPU architecture,which is used to find the optimal algorithm parameters suitable for different hardware and different scale problems.Firstly,we analyze the bottleneck of the basic zero-knowledge proof implementation,find the key tasks,and establish the key task performance analysis model,which is divided into three parts:CPU computing time,CPU-GPU transmission time,and GPU computing time.After testing the small amount of data,we can determine the coefficient related to the hardware environment in the model,and then use the coefficient to predict the performance of large-scale data input in real application scenarios,and find out the optimal value of adjustable parameters in the algorithm.Then,based on the basic implementation of zero-knowledge proof,we added multi-stream mechanism for optimization,which has the effect of hiding part of CPU-GPU transmission time.Based on the key task performance analysis model,a performance analysis model of multi-stream mechanism optimization scheme is proposed to find the best stream number.Finally we combine the previous three parts of the work,put forward the zero knowledge proof optimization method.After analyzing the bottleneck of the basic zero-knowledge proof implementation,we put forward three feasible optimization ideas.For the key task performance analysis model,we used this method to test it on two servers with different gpus.The input data for the test were real data from Filecoin project and test data Dummy from bellperson.The results show that our model can achieve 89.5% and 85% accuracy on Ge Force RTX 2080 TI and Ge Force RTX 3080,and performances are optimized by 33.5% and21.5% compared to the average performance.For the performance analysis model of multi-stream mechanism optimization method,we tested it on Ge Force RTX 2080 TI and achieved 90% accuracy.And when the input data is the same as the original implementation,compared with the original implementation after parameter tuning,the optimization effect is 7%.For the zero knowledge proof optimization method,we get73% optimization effect on the premise of ensuring the correctness of proof.