Not backing up will cost you!

[superboyac]

I see.  Yes, i'm probably misusing the term.  I figured JBOD meant exactly what I thought, but it's not.  What's the difference between what I'm describing and Raid-1?

[4wd]

I see.  Yes, i'm probably misusing the term.  I figured JBOD meant exactly what I thought, but it's not.

-superboyac (December 30, 2010, 06:34 PM)

No, it might just be me.    As I said, generally JBOD has implied what is really meant as SPAN when people talked about it.  But JBOD really means what you have stated, Just a Bunch Of Drives - each individually addressable.

So your use is correct, however specifying JBOD still implies no data redundancy.

What's the difference between what I'm describing and Raid-1?

With RAID-1, one HDD is mirrored to another - they both contain exactly the same data, ergo they, (AFAIK), have to be the same capacity.  It's also taken care of by the controller, no user interaction, (apart from initial setup), required unless one HDD dies.

File syncing requires that the user run additional software, (and makes sure it's running), and action any errors, eg. files locked, permissions, etc, however the HDDs can be different capacities or the same HDD but different partition.

Incidentally, following on from worstje and JJ above, if you were going to try RAID-1, (even if only on a pair of your HDDs), I'd suggest using two different manufacturers, eg. WD 2TB and a Samsung 2TB, you'd be unlikely to end up with two HDD from the same batch.  Also, it would probably be better to use Enterprise class HDDs instead of Desktop class.

[40hz]

Generally, JBOD refers to combining separate HDDs into effectively one bigger HDD so that data is spanned across all of them.  So if the 3rd HDD of a 5 HDD array dies, you've generally lost all data across all HDDs.  You may be able to recover files using recovery software from the other HDDs but it'll be a long process.

By definition, JBOD is non-RAID and therefore there is no data redundancy.

-4wd (December 30, 2010, 05:33 PM)

Actually, aren't there are two definitions for JBOD? One where data is spanned across the disks as you describe; and the other where a virtual disk manager is put on top of the "bunch of drives" but where each physical disk operates independently of the others?

If I recall correctly Goolge's G. Boudreau's Greyhole Project utilizes the second approach, but goes it one better by providing for data replication in software. Doing that makes Greyhole act much like a RAID-5 combined with an extremely flexible drive concatenation capability. This is similar to the 'Drive Extender' feature Microsoft just removed from their new Home Server release.

Last I heard it was still pre-1.0 release. But it is working and some people are using it.

Might be worth a look. (Sorry, don't have the link right this moment, but it's easy to find by - surprise! surprise! - googling 'Greyhole Project.') Link here.

------------
ADDENDUM: Whoops! 4wd got in there just ahead of me!  


EDIT: I stand corrected. Greyhole isn't Google's. It's the work of Guillaume Boudreau. Project website here. Personal blog here.

 

[4wd]

Actually, aren't there are two definitions for JBOD? One where data is spanned across the disks as you describe; and the other where a virtual disk manager is put on top of the "bunch of drives" but where each physical disk operates independently of the others?

-40hz (December 30, 2010, 07:03 PM)

The spanned version is more properly called SPAN (DOH!) or BIG now, with JBOD correctly being just individually addressed drives.  However, even some company descriptions seem to get it wrong.

eg. Some of Addonics products refer to JBOD (Concatenation), which I would have thought would really be SPAN or BIG, which they refer to on others.  Some have JBOD (Individual) and BIG.

If I recall correctly Goolge's Greyhole Project utilizes the second approach, but goes it one better by providing for data replication in software. Doing that makes Greyhole act much like a RAID-5 combined with an extremely flexible drive concatenation capability. This is similar to the 'Drive Extender' feature Microsoft just removed from their new Home Server release.

That looks very interesting, kind of like a combination of AFS, ZFS and RAID-5 

[superboyac]

This is something I've been meaning to figure out, so I'm glad we're talking about it here.  This stuff is confusing, complicated, and has lots of terms.  So I'm going to try to simplify for myself.

In my situation, the storage is more so the goal rather than performance.  I say that because I'd like to keep this as simple as possible.  What I don't like about raid is the hardware requirements such as the disks having to be the same size.  I mean, it makes sense that a backup drive is the same size, I just don't like it to be a requirement.  I'd like to just cram in whatever drives I happen to find lying around.  The synchronization and all that stuff will be taken care of with software, so once again, I'd like to avoid using hardware for things software can do.  The only advantage I see in using hardware is performance.  As long as I can copy files back and forth and streamp 1080p, that's all the performance I need.

Unless I am not understanding something, I'll be staying away from RAID.  I'll be using a bunch of individual drives.  I'll use Windows 7 library tools to create folders that span multiple drives.  I'll use Super Flexible to sync my files back and forth (which I already use).  I think that's a safe, solid setup, no?

I like these other spanning ideas mentioned above by 40 and 4wd, but unless I need additional features, I'm going to keep it real simple.

I do have a question, how do all these individual drives connect to the server?  I can't imagine there's 16 SATA cables coming out the back plugging into the motherboard?  Does it use a CAT5?  Is something going on inside the box which combines it into one SATA or eSATA cable?  How does this work?

[4wd]

I do have a question, how do all these individual drives connect to the server?  I can't imagine there's 16 SATA cables coming out the back plugging into the motherboard?  Does it use a CAT5?  Is something going on inside the box which combines it into one SATA or eSATA cable?  How does this work?

-superboyac (December 30, 2010, 07:56 PM)

Port Multipliers - one eSATA port into 4 or 5 SATA drives, or Infiniband - 4 SATA ports into one cable and then back to 4 SATA drives, (maybe even 20 drives if you can tack Port Multipliers on the end).

Addonics, (yes, them again ), sell Port Multipliers.  The 5-Port HPM-XA on a PCI bracket will give you access to 5 SATA drives over 1 eSATA cable.

If you will be going to move most of the drives out of your system box into an external enclosure then that will free up more internal SATA connectors to be used as eSATA ports or a couple of Infiniband ports.

The Norco case you seem to have chosen allows for the installation of a motherboard and since you're going to connect the drives to a network for house-wide access which would require a PC anyway, (or some form of SATA->Network interface), then you may as well install a motherboard and create a NAS, (using whatever OS tickles your fancy - XP Pro, WHS, etc for simplicity of filesync software or eg. FreeNAS for full-blown NAS).  This will give you lots of SATA ports, (whether onboard or PCIe cards), and the Gb interface.

Addendum: If you were going to go for the NAS thing, then possibly a motherboard with multiple Gb ports that allows 'Teaming', (ASUS term, Gigabyte use something else), this would reduce any bottleneck from the network.

[MerleOne]

So here's part 1 of 2 blog posts I'm writing about my data loss, recovery, and backup experiences. I hope it will serve as a stern reminder to everyone to backup their important data. Not doing so could cost you 1000's of dollars or worse.

http://oshyan.blogsp...ck-up-your-data.html

- Oshyan

-JavaJones (December 24, 2010, 05:34 PM)

Pretty interesting article.  Thanks !  I wonder how a Drobo would have performed there.

[Stoic Joker]

RAID isn't a bad thing, it's just not something you can do safely in a half-assed manner. If you're seriously looking at building a box to permanently store 10TB of data - that you actually expect to have 5-10 years from now - Do Not Frankenstein it together from "kit" parts. Stuff breaks, but never in a timely fashion ... You need to get something mainstream. From a major manufacture ... That will have replacement parts available down the road.

You can get (refurbished) off-lease commercial hardware for a fraction of the initial selling price. And reputable resellers will also provide warranties for up-to a full year. As an example:

 Here is a Dell PowerEdge 2950 for $860. It's a Dual Xeon with 4GB RAM, and a True Hardware RAID controller which will handle up to 6 2TB SATA drives. Add a completely legal server license for an additional $250.

 I have personally bought several refurb servers from them for both in-house and client use, and their service has always been excellent.

[Stoic Joker]

There is a section on my site in the Quick Facts section about RAID types and usages. It was made up (long ago) of reproduced articles that I didn't want to lose so I could refer to them quickly if need be.

RAID Configurations

[JavaJones]

IMHO RAID (how's that for acronyms? ) is simply *not* useful for 99% of normal, home (or home office) users. The only real benefit is speed, and you need a normal backup solution anyway, so unless you have high availability requirements for the data, there is no need to consider RAID as a "safety system".

The unit I was using is built in-whole by a major manufacturer (Lacie), one that many high end photographers, small-to-medium businesses, and more trust with their data. I do *not* trust them anymore, in part because of their lackluster support, in part because of the higher-than-I-think-is-reasonable failure rate of their equipment (see Amazon reviews of 4big).

All that being said yes, if you're going to do RAID, do it right. It aint cheap to do it right, making it all the more clear to me that for most home or small office users it is neither necessary nor useful, and may even be dangerous.

Still working on part 2 of the blog post. I had an interesting opportunity to test my current backup strategy just yesterday, and it worked, thank god.

- Oshyan

[f0dder]

I do striping (which really shouldn't be called RAID, since it's anti-redundant) on my workstation, because I like the added speed, and only use that partition for relatively volatile stuff - and one big partition is more useful than two 74gig partitions (yes, I'm striping my raptor drivers). Definitely useful.

I mirror my fileserver, and I find this extremely useful as well - it unfortunately doesn't do hot-swap, but five minutes power-down and HDD swapping is a lot better downtime than having to restore a new drive from backup. As long as you don't use mirroring instead of backups.

But JBOD, RAID-5 or other solutions? Wouldn't even consider it.

[JavaJones]

Exactly.

- Oshyan

[superboyac]

You know, every time I come here and read the discussions about backing up, I continue to get confused.  you guys say stuff like RAID isn't for backup, and I get that.  Then you say something like:

But JBOD, RAID-5 or other solutions? Wouldn't even consider it.

And I don't quite understand what that means, really.  What's wrong with JBOD?  Why can't I just use a bunch of disks and have the files and folders synced with another duplicate drive?  So each drive has a backup drive that simply copies the files/folders.  No fancy mirroring or indexing.  Just straight up copying/moving/deleting...basic file operations.  And I can yank any drive out at anytime, plug it into another computer and see those files/folders without any extra steps.  What's wrong with that?  What have I left unaccounted for?

I'm trying to come up with the ultimate backing up solution...for HOME users.  A home user with a lot of data, yes, but still just a home user.  I want to make sure I have all of my files safeguarded against just about any situation.  If that means double or even triple redundancy, so be it.  If it means imaging or versioning, so be it.  But i always get confused whenever one of you says something like "Oh my gosh!  I would NEVER use file syncing as a backup method."  I'm like, "Why the hell not, I'm doing that!  Shit!  What's happening?!"

Anyway...

[f0dder]

Superboy, JBOD means making a bunch of (possibly irregularly sized) drives appear as one large drive - which is pretty bad for your data if even a single drive crashes (although not as bad as striping, where your data is evenly distributed across all drives). JBOD is a raid mode, not "just stuffing a bunch of drives into a single computer".

[superboyac]

Superboy, JBOD means making a bunch of (possibly irregularly sized) drives appear as one large drive - which is pretty bad for your data if even a single drive crashes (although not as bad as striping, where your data is evenly distributed across all drives). JBOD is a raid mode, not "just stuffing a bunch of drives into a single computer".

-f0dder (January 04, 2011, 05:04 PM)

OK, then, we keep getting confused on this JBOD.  Apparently, JBOD doesn't actually mean "Just a bunch of disks" to you guys.  When I say JBOD, I mean exactly what you jsut said: stuffing a bunch of random drives onto one system.  What is the term for that, if it's not JBOD, because I'm going to be using it a lot this year.  I don't want to keep confusing people with thinking I'm trying anything that has to do with RAID, or anything more than simply having files/folders on a disk drive.

[JavaJones]

I don't think JBOD should really be used in the context f0dder intends (although it *is* common usage to do this, which has created the confusion). SPAN or BIG are perhaps more appropriate but also more specific. The general term is concatenation. http://en.wikipedia....e_architectures#JBOD

- Oshyan

[f0dder]

I don't think JBOD should really be used in the context f0dder intends (although it *is* common usage to do this, which has created the confusion). SPAN or BIG are perhaps more appropriate but also more specific. The general term is concatenation. http://en.wikipedia....e_architectures#JBOD

-JavaJones (January 04, 2011, 05:24 PM)

Confusing, perhaps... but since it tends to be on the feature list on just about any half-assed "hardware raid" card as well as 100% software based solutions, and has been for quite a lot of years, you definitely shouldn't say "JBOD" when you really do mean Just a Bunch Of Disks

Btw, doing (windows/linux/whatever) application-based syncing between a bunch of drives in a single server sees of relatively limited use, unless you're also doing versioning. For just syncing, mirroring would be better (a purely software-based solution is fine, really, but you do want OS-level support, or at least something driver-based).