JBOD: reliability VS single disks

[Loa]

Hello,

I have two 2TB disks in a two drive enclosure, and one of them is running out of space. I know I could use the other, but I prefer to have similar items in the same logical folder.

My question is this: is JBOD less reliable than single disks on their own? From what I understand it should be just as reliable, but I wanted to make sure.

Also, can I form a JBOD without losing the data that's already on the disks?

Thanks

Loa

 

[nanofrog]

I have two 2TB disks in a two drive enclosure, and one of them is running out of space. I know I could use the other, but I prefer to have similar items in the same logical folder.

My question is this: is JBOD less reliable than single disks on their own? From what I understand it should be just as reliable, but I wanted to make sure.

It's basically the same (one disk fails, the others still contain data), unlike a stripe set, where a single failure will destroy all the data.

But please note, that in 10.4, a failure of a JBOD would also cause a total data loss. I presume this has since been resolved. (Source = Wiki).

No one's listed such a failure, and is where the presumption comes from (rather stupid to continue to carry that behavior over from revision to revision IMO).

Also, can I form a JBOD without losing the data that's already on the disks?

Yes.

That's one of the reasons why it's easier to create than say a stripe set. No need to move the data off of the disks intended as members, create the set, then restore (unlike a stripe set or true RAID level, which has an initialization stage that will wipe out the data).

Expansion is also easy; just add a disk/s to the set. Data will remain.

 

[Loa]

Thx for the info.

I'm a bit scared of the creation of the JBOD, because while I do have back-ups, they're on DVDs. Re-copying 2TB of stuff from DVDs isn't a fun process!

Loa

 

[nanofrog]

I'm a bit scared of the creation of the JBOD, because while I do have back-ups, they're on DVDs. Re-copying 2TB of stuff from DVDs isn't a fun process!

Hmm... let me see... more than 400 DVD's... No, I wouldn't think it fun at all.

 

[deconstruct60]

My question is this: is JBOD less reliable than single disks on their own?

JBOD being a bunch of disks all presented as separate volumes? Then no as long as look at them as individual disks/volumes. If want to only consider them as a logical volume then see next entry. There JBOD is really just added another disk controller to the box.

JBOD being merged disk spaced to present as one single logical volume. Then yes.

Perhaps not a big enough difference to put you off though. JBOD to create a single volume has the same failure/reliability issues that RAID 0 has:

http://en.wikipedia.org/wiki/Standard_RAID_levels#RAID_0_failure_rate


In general, adding more components to a system increases the change of a failure. Most of the upper levels of RAID are about whether the loss is catastrophic or not. The arrival of failures goes up though as add nodes though.

A JBOD that survived a failure would either :

need a special file system which redundantly stored meta data (so only hierarchy is always around no matter what)

implement some pseudo RAID-1 replication of all data blocks.

or

be something chessy like a virtual symlink of several seperate volumes to look like there were just one volume.

fakevol /
symlink to disk1
symlink to disk 2
symlink to disk 3

each disk had its own filesystem hierarchy. Only really buying is less disk volumes mounted on the desktop.

options 1 and 3 are still failures. Just not catastrophic to extent filesystem cannot is crushed to point can't tell you there was a failure.

 

[Loa]

Interesting info deconstruct, but what's stopping me right now is that Apple's DU implementation of the JBOD (Concatenated) destroys the data on the disks.

Loa

 

[nanofrog]

Interesting info deconstruct, but what's stopping me right now is that Apple's DU implementation of the JBOD (Concatenated) destroys the data on the disks.

Loa

I don't recall reading/seeing that this was the case in creating the set, but 10.4 did have problems with recovery (all data lost). I've presumed that the recovery issue's been sorted by now, as that's a bad implementation if it does wipe the disks. Worse, if creation or expansion also wipes data (no way to easily expand a set if this is the case, as a JBOD can very easily out-size any other storage pool).

By chance, how did you come by this information, and is newer information on recovery out (i.e. 10.5 and 10.6 on)?

Just another reason I'm not a fan of software implementations (by far prefer hardware based implementations), but I've usually not had the number of issues with JBOD as with RAID levels for software implementations. But there are those few exceptions (poor implementations), that can cause all kinds of headaches.

I've just not tested JBOD with OS X's Disk Utility, and it's no longer possible on a MP, as I returned it (2008).

 

[Loa]

By chance, how did you come by this information

From the horse's mouth: it's a warning in Disk Utility just before you start the process.

and is newer information on recovery out (i.e. 10.5 and 10.6 on)?

I haven't been able to find anything new.

Loa

 

[nanofrog]

From the horse's mouth: it's a warning in Disk Utility just before you start the process.

Ouch.

I haven't been able to find anything new.

Between this (unknown behavior), and the fact that the data will be wiped during the creation of the set, I'd look to something else if it were me.

Usually you could go for a simple HBA or SATA card, but as it's a MP, you're in a tight spot, as you'd never be able to access the card's firmware in the MP.

This means a RAID card, as you have to be able to access the card's settings through either a utility or a browser (proxy server). Areca and ATTO would do for brands, and you can boot from them if needed, by flashing the boot section of the firmware to EFI on the card (done through the utility or browser, depending on brand chosen). Areca = browser access, ATTO = utility.

Physical installation may be a bit challenging (i.e. more $$$ involved for adapters and cables for internal, and cables and enclosures for external, which I presume is the route intended).

The easiest and least expensive would be an ARC-1210 (SATA connnectors). Don't bother with Highpoint, as their support section sucks.