| 网络并行计算中的远程直接内存存取实现 |
| |
| 引用本文： | 乔崇,张俊杰,武杰,刘尉悦,王砚方.网络并行计算中的远程直接内存存取实现[J].中国科学技术大学学报,2006,36(3):273-280. |
| |
| 作者姓名： | 乔崇 张俊杰 武杰 刘尉悦 王砚方 |
| |
| 作者单位： | 中国科学技术大学近代物理系快电子学实验室,安徽,合肥,230026 |
| |
| 摘 要： | 提出了一种基于硬件TCP协议（HTCP）的远程直接内存存取（remote direct memory access,RDMA）设计.该设计直接架构在网络的MAC协议层上,因此仅用非常简单的硬件就可实现较高的传输效率,适用于不需要穿越网关的小规模并行计算.实测结果表明使用这种HTCP协议的RDMA网络传输对CPU的占用率比采用普通的TCP协议至少降低一半以上.
|
| 关 键 词： | 远程直接内存存取 并行计算 |
| 文章编号： | 0253-2778(2006)03-0273-08 |
| 收稿时间： | 11 8 2004 12:00AM |
| 修稿时间： | 10 9 2005 12:00AM |
A new realization of RDMA for network parallel computation |
| |
| QIAO Chong,ZHANG Jun-jie,WU Jie,LIU Wei-yue,WANG Yan-fang.A new realization of RDMA for network parallel computation[J].Journal of University of Science and Technology of China,2006,36(3):273-280. |
| |
| Authors: | QIAO Chong ZHANG Jun-jie WU Jie LIU Wei-yue WANG Yan-fang |
| |
| Institution: | Fast Electronic Lab, Department of Modern Physics, University of Science and Technology of China, Hefei 230026 ,China |
| |
| Abstract: | A remote direct memory access(RDMA) design based on hardware TCP(HTCP) was proposed.The design is built directly on the MAC protocol layer of the network,so high efficiency and can be easily realized with simple hardware.It is applicable for small scale parallel computation areas in the same gateway.The result of test shows that after using HTCP in RDMA network transmission,the CPU usage is less than half of using normal TCP. |
| |
| Keywords: | FPGA PCI |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
