摘要
为了提高区块链中节点的验签效率,提出基于GPU的区块链交易验签加速技术.结合CPU-GPU异构平台架构特性对交易验签过程进行分阶段优化,大幅提高SM2验签算法运行效率,同时充分利用GPU内核调用的异步性,有效降低交易验签过程的整体IO开销.考虑到GPU计算能力强而分支预测能力弱的特性,提出改进的同时多点乘算法,不仅提升了GPU验签效率,而且增加了多线程并行规模.所提方法将交易验签操作卸载至GPU处理,释放了节点被占用的CPU资源,在不修改区块链协议的情况下实现了区块链系统整体性能的提升.基于RTX3080平台和国产许可区块链Hyperchain平台进行实验,结果表明,该方法峰值验签吞吐量为4.52×10^(6)次/s,集成该方法的Hyperchain平台交易吞吐量提高了15.81%,且延迟下降了6.56%.
A GPU-based acceleration technology for signature verification of blockchain transactions was proposed,in order to improve the signature verification efficiency of peers.A phased optimization for signature verification process of blockchain transactions was performed by combining the characteristics of CPU-GPU heterogeneous platform architecture,greatly improving the efficiency of the SM2 verification algorithm.The asynchronous nature of GPU kernel calls was fully utilized,effectivelly reducing the overall IO overhead of the transaction signature verification process.Considering the characteristics of GPU with strong computational power and weak branch prediction,an improved simultaneous multipoint multiplication algorithm was proposed,which not only improved the efficiency of GPU signature verification,but also increased the multi-threaded parallelism scale.The proposed method offloaded the transaction signature verification operation to GPU processing,which freed up the CPU resources of peers occupied and improved the overall performance of the blockchain system without modifying the blockchain protocols.Experimental results based on the RTX3080 platform and Hyperchain,a domestic permissioned blockchain,showed that the peak signature verification throughput of the proposed method was 4.52×10^(6) transactions per second,and the transaction throughput of Hyperchain platform integrated with the proposed method increased by 15.81%while latency decreased by 6.56%.
作者
崔璨
杨小虎
邱炜伟
黄方蕾
CUI Can;YANG Xiao-hu;QIU Wei-wei;HUANG Fang-lei(College of Computer Science and Technology,Zhejiang University,Hangzhou 310027,China;Hangzhou Qulian Technology Co.Ltd,Hangzhou 310000,China)
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2023年第8期1505-1515,共11页
Journal of Zhejiang University：Engineering Science
基金
浙江省科技计划资助项目(2022C01126)
国家重点研发计划资助项目(2021YFB2701100)。
关键词
区块链
交易验签
GPU加速
吞吐量
延迟
blockchain
transaction signature verification
GPU acceleration
throughput
latency
