| 并行文件系统的设计 |
| |
| 引用本文： | 孙凝晖.并行文件系统的设计[J].计算机学报,1994,17(12):938-945. |
| |
| 作者姓名： | 孙凝晖 |
| |
| 作者单位： | 国家智能计算机研究开发中心，中国科学院计算技术研究所 |
| |
| 摘 要： | 在大规模并行处理巨型机（ＭＰＰ）的设计中，提高Ｉ／Ｏ性能同提高计算能力和通信能力同样重要。并行文件系统（ＰＦＳ）在多个Ｉ／Ｏ结点的多个磁盘上，分布文件系统和文件的磁盘块，将文件读写在计算结点转化成多个对物理块的直接Ｉ／Ｏ请求，利用预读，预分配，磁盘缓冲式区和异步Ｉ／Ｏ增加Ｉ／Ｏ的并发生，在特定的文件使用模式下，也是ＭＰＰ应用的主要Ｉ／Ｏ模式，获得很高的Ｉ／Ｏ效率。
|
| 关 键 词： | 巨型计算机 并行文件系统 磁盘缓冲 设计 |
THE DESIGN OF PARALLEL FILE SYSTEMS |
| |
| Sun Ninghui.THE DESIGN OF PARALLEL FILE SYSTEMS[J].Chinese Journal of Computers,1994,17(12):938-945. |
| |
| Authors: | Sun Ninghui |
| |
| Abstract: | In massively parallel processing supercomputers, the efficiency of I/O is as important as computing power and communication bandwidth. Parallel file system distributes file system and disk physical blocks of a single file among several disks on several I/O nodes. PFS translates file read/write into direct I/O requests for physical blocks in computing nodes, and improves the I/O.parallelism by preread, pre-allocation, disk cache and asynchronous I/O. In the main I/O model of MPP applications, the I/O effeciency of PFS is better than conventional file systems. |
| |
| Keywords: | Supercomputer parallel file system disk cache pre-read preallocation |
| 本文献已被 CNKI 维普 等数据库收录！ |
