Does cluster size really matter?
i have a db server that has 4k cluster sizes on an NTFS volue. THe log fies are on another volume (NTFS). both drives are striped, and mirrored over 3 disks in the stripe set.
i ran perf mon. it’s not cpu. it’s not memory. DISK utilization is over 80%.
the system is SLOW! for a quad processor box with 4gb ram.
the cpu hardly moves (30% or less). Page faults are high, but the weird thing is that available ram is 2gb free.
i have the server set for background applications, and the db server to use all available memory as needed.
I ran HDTach. the score shows 114mbps random transfer on the raid stripe set.
i’m guessing either the DB iteslf is porrly designed OR the File I/O is so fragmented, and the cluster sizes are too small.
so that’s my question: Does Cluster size matter? i don’t mean theory or i think so. does any one have a document pointing to actual performance tests on disk i/o and cluster size?