[TriLUG] Linux Software RAID experience

[Jim Salter via TriLUG]

If you want a completely hands off "just stick the drive in and go" 
experience, you're going to be stuck with hardware RAID. It's /easy/ to 
rebuild arrays with either mdadm or zfs, and the CLI work can be done 
remotely after the drive is changed, but they are not "yank it and stick 
it and forget it".

IME mdadm will generally beat the pants off most if not all hardware 
RAID. But honestly, the workload you're asking for - sequential reads - 
is easy for anything; it does sound like you've likely got a hardware 
issue there.

I do question the need for sequential throughput rather than IOPS if 
this is an NFS server for a virt cluster, though. Almost any VM 
operations are heavily IOPS-dependent.

One other thing you might want to make sure of is that you remove the 
NFS from the equation when testing. Are you testing on the local server, 
or across the NFS mount?