An efficient caching scheme for software RAID file systems inworkstation clusters

摘要

A software RAID file system is defined as a system thatdistributes data redundantly across an array of disks attached to eachof the workstations connected on a high speed network. This provideshigh throughput as well as higher availability. In this paper we presentan efficient caching scheme for the software RAID file system. Theperformance of this scheme is compared to two other schemes previouslyproposed for conventional file systems and adapted for the software RAIDfile system. As in hardware RAID systems, we found small-writes to bethe performance bottleneck in software RAID file systems. To tackle thisproblem, we logically divided the cache into two levels. By keeping olddata and parity values in the second-level cache we were able toeliminate much of the extra disk reads and writes necessary forwrite-back of dirty blocks. Using trace driven simulations we show thatthe proposed scheme improves performance by at least 10% for both thehit ratio and the average system busy time