并行文件系统自适应的文件条带化技术

Automatic-adaptive File Striping of Parallel File System

研究并行文件系统自适应的文件条带(Striping)策略对改进文件访问性能的影响,并开发动态的文件条带分析模型,利用自动访问模式分类和实时文件系统性能数据为文件条带策略选择模糊逻辑规则库,来优化文件访问性能。研究结果表明:当文件系统负载低时,可以尽量将文件分布到所有磁盘上来最小化I/O的反馈时间;反之,在系统负载高时,使文件分布的范围小一些以便最大化文件系统整体的吞吐量。并通过实验给出了请求大小、请求宽度、请求到达率与系统性能的相互关系,实证了自适应规则库的正确性。

This paper studies a fuzzy logic rule base for adaptive striping of files across multiple disks, and the rule base is based on an analytical model of disk contention that includes disk physical parameters and file request sizes; based on auto classifying of access patterns and real-time monitored data of file system. As the file system load is low, the rule base stripes files aggressively to minimize response time. At high loads, it stripes less aggressively to maximize aggregate throughput. Experiments results show how do request size, request width, request arrival rate affect I/O performance, and verify the correctness of auto-adaptive rule bases.