Hard Disk device IDs in Linux

Lonnie Olson lists at kittypee.com
Fri Mar 8 09:18:23 MST 2013

On Fri, Mar 8, 2013 at 8:49 AM, Michael Torrie <torriem at gmail.com> wrote:
> Anyway, what the device nodes are is completely up to the distro.  When
> I managed RH servers connected to a big san, we didn't use
> /dev/sda-style notation at all.  Instead we used /dev/disk/by-id/ nodes.
>  This way even if the disks moved to a new chassis, they'd still be
> found.  We did find, though, that the RAID chassis itself would create
> these ids, and when you moved a RAID set to a new chassis, the disk ids
> would change.  But it was much better than using /dev/sda and the like,
> which could change order at boot, depending on how fast the arrays
> responded and were enumerated by udev.

I personally like using the UUID= or LABEL= methods for identifying
disks to mount.  But anyway your idea to use a different identifier
other than /sev/sda-style is a good one for a machine where disks
might move around.

I would not recommend using that many drives in a single RAID array.
The more disks you have in a single RAID6 array increases the chances
you will encounter multiple disk failures and hence a whole array
failure.  Personally I wouldn't use more than 10 drives per RAID6, but
this number can of course vary.  If you want them all in a single
array, you can group the individual RAID6 arrays into a single one by
concatenating or striping the arrays.

More information about the PLUG mailing list