| 安述倩* **,李文明*,范志华* **,吴海彬*,吴萌*,王达*,张浩*,唐志敏* **.NDP-Ledger：面向区块链应用的通用高通量加速架构[J].高技术通讯(中文),2020,30(11):1093~1103 |
| NDP-Ledger：面向区块链应用的通用高通量加速架构 |
|
| |
| DOI：10.3772/j.issn.1002-0470.2020.11.001 |
| 中文关键词: 区块链； 3D堆叠存储； 近数据处理(NDP)； 数据流； 通用加速器结构 |
| 英文关键词: blockchain, 3D stacked memory, near-data-processing (NDP), dataflow, general-purpose acceleration architecture |
| 基金项目: |
| 作者 | 单位 | | 安述倩* ** | | | 李文明* | | | 范志华* ** | | | 吴海彬* | | | 吴萌* | | | 王达* | | | 张浩* | | | 唐志敏* ** | |
|
| 摘要点击次数: 2612 |
| 全文下载次数: 1599 |
| 中文摘要: |
| 区块链技术由于去中心化及不可篡改等特性，广泛应用于数字货币、支付交易等领域，其算法对计算能力和存储访问能力有较高要求，导致传统冯诺依曼结构在面向区块链应用时能效比较低。3D堆叠存储器因可以缓解冯诺依曼结构的访存瓶颈，成为了热门研究方向之一。本文基于3D堆叠存储器技术及数据流执行模式，提出了一种面向区块链应用的高通量近数据处理（NDP）架构，NDP-Ledger。本文深入分析和论证了区块链应用的计算特征及3D堆叠存储技术在区块链应用中的适应性问题，并基于数据流执行模式设计了一种通用的高并发区块链加速体系结构，使区块链加速器在满足通用性的前提下提高处理性能。模拟评估结果表明，本文提出的区块链通用加速器结构，在典型区块链应用处理方面的性能优于主流的CPU和GPU。 |
| 英文摘要: |
| Blockchain technology is widely used in the fields of digital currency, payment due to its features of decentralization and tamper-proof. The algorithms require high computing power and storage access capacity, resulting in low energy efficiency of traditional von Neumann structures in blockchain applications. The 3D stacked memory, which alleviates the bottleneck of von Neumann structure, has become one of the hot research fields. Based on the 3D stacked memory technology and data flow execution mode, a high-throughput near-data-processing (NDP) architecture for blockchain applications (NDP-Ledger) is proposeed. It analyzes and demonstrates the computing characteristics of blockchain applications and the adaptability of 3D stacked storage technology in blockchain applications. A general-purpose high-concurrency blockchain acceleration architecture is designed based on the dataflow execution mode, which can improve the processing performance on the premise of meeting the universality. The evaluation results show that the general-purpose acceleration architecture of the blockchain proposed has better performance in typical blockchain application processing than CPUs and GPUs. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
