摘要

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.

全文