(For the archives:)
The "RAID-1 dual archival backups" solution is working, but we
needed to do some tweaks.
First, we tried the built-in RAID-1 feature that came with an
external enclosure. This basically worked, but had some major usability
problems:
1. When you slap in a new disk set, it immediately starts rebuilding the
RAID-1 array. Since the hardware enclosure is ignorant of all
partitions and/or filesystems, it is not able to detect that you have
two brand new (blank) disks that don't need to mirror. So, even with
two new disks, you need to wait several hours for the RAID-1 to
initialize the entire disk. (We are using 1.5TB disks.)
2. The RAID-1 array is specific to the particular hardware device. This
is really only a problem if you want to re-create the array, but I have
a personal hatred of all hardware-specific RAID systems.
3. The particular enclosure we used was an ICY DOCK. It's nice because
it requires no screws or caddy. Just slap the disks in and go. But
it's a problem because the left slot is always considered the "MASTER"
of any RAID set. So if you modify one of the RAID-1 disks, and put it
in the wrong slot with a new disk, it will wipe your modifications and
create a blank RAID-1 set.
Because of these issues, we decided to use a software-based RAID
system instead. This was in a Mac environment -- the Macs have
something just like mdraid (except that it only supports RAID-0 or
RAID-1.... lol, Macs are sooo cute).
By using the software RAID, we have a solution that will work on
any of the Mac workstations. It's not bound to any hardware. We can
even use two separate single-disk USB enclosures. It's all the benefits
of using mdraid, but for the Macs. The resulting disks are readable as
standalone, non-RAID single disks. When you create a new array (on two
brand new, blank disks) it does not try to initialize them, it just
starts mirroring them from then on (so it's immediately ready to go).
To make this work we had to flip some dip switches on the ICY DOCK
to put it into JBOD mode. We were able to take the disks, put them into
a totally different JBOD enclosure, and have the RAID set recognized.
In short: if buying a multi-disk enclosure, make sure it supports JBOD.
We are still investigating a possible compatibility issue on a
newer Mac (with a newer Mac O.S.), but in general, it seems to be
working. (I have no idea if Linux can recognize a Mac RAID set, we
haven't tried.)
Thanks for all advice and assistance!
--Derek Simkowiak
On 10/28/2010 08:38 AM, Brian C. Lane wrote:
On Mon, Oct 25, 2010 at 04:35:46PM -0700, Derek Simkowiak wrote:
But I suspect that plan will fail because the RAID-1 disks will
be useless to anyone who does not have the specific make-and-model
of eSATA drive case. For example, an LSI card will not recognize a
3ware RAID set on a batch of disks, so I suspect that a simple
(non-RAID) SATA controller on a mobo will not be able to do anything
with a single disk of the RAID-1 setup mentioned above.
Or am I wrong because RAID-1 does no striping, just simple
mirroring? Does anyone have experience with this?
If you use mdraid with v1.0 (or earlier) metadata the partition will
look just like an ext4 filesystem because the metadata is written to the
end. v1.1 and newer write to the front of the partition so it isn't
obvious what fs is on it.
If you're using Linux to read the backups this really doesn't matter, it
will recognize either. Unless speed is really an issue I'd use mdraid so
that you aren't tied to any specific RAID hardware. Otherwise, make sure
you have spares :)
I expect the biggest hurdle, with mdraid or hardware, is going to be
speed, given the size of the dataset.
Brian
