Hi all,
We've finally (!) found time to get our DPT SmartRAID V up & running. We've
got 9 x 18 GB disks (LVD => fast; it turns out our other 8 disks are not LVD
and I feel kinda dumb for buying them) active in RAID-5.
Now I'm in the process of making a big ext2fs on the DPT raid, and the file
system is not quite as quick as I'd have hoped. I suspect it's got something
to do with our settings for ext2's raid ``stride'' option (since nothing
else seems relevant but I'm not sure what the stride option does, and the ext2fs docs
escape my sight or are opaque.
So here's what we're after: we've got a lot of moderately large files and our access
patterns hit a lot of them at once. We'd rather acheive speedups through multiple
concurrent file transfers than through striping, so we'd like large blocks (or
clusters of blocks) contiguous on each disk. This is what we'd achieved under Linux's
software raid5 where we interleaved at 1MB boundaries, but the DPT controller only
allows clusters that add up to 1 MB (which means 64kB/disk in our case).
Our most recent mke2fs used the following parameters:
blocksize=4096
stride=16
Are these parameters appropriate?
Do you have suggestions on how to improve things?
Thanks,
-- Josh Fishman and Eric Freudenthal
NYU / RLab
