Is a home NAS reliable to keep your data on (no other backups)?

Question

The question came up while I was thinking to move some data (e.g.) pictures, old but important other data solely to the NAS (private use, no business level).

My NAS is a simple 2 disk RAID system. I know that a RAID is not a backup system as it only provides hardware redundancy so I planned to create incremental (versioned) backups of the data directories. The backup destination would be the same NAS so that the only instance of the data (no copies on notebook or desktop machines) and the versioned backups would stay entirely on those two disks.

Now I wondered how reliable this procedure is. I've read that a power blackout has typically no effect on the data on a disk, as the heads retract to a safe position and do not "hit" or "scratch" the disks, but it can of course corrupt data currently written and even worse affect the electronics of a drive rendering it unusable.

Now I am pretty unsure whether I should invest in a small UPS that plays well with the NAS to ensure proper shutdown or resort to multiple backup locations (or both?)

Does having only one instance of data + backup on same machine still regarded unsafe?

(I am ignoring theft here. Also surges, as all connected cables LAN, Power etc. are protected. Even more: I do not want to use the cloud for any data (encrypted or not))

Edit: Sorry, it is a RAID 1 with two disks total.

Solution 1

Is a home NAS reliable to keep your [old but important, private use, no copies on other machines] data on (no other backups)?

Noooo!

Don't even think about doing this, unless you fancy losing your data. Since you refer to the data you intend to store there as "important", I imagine you don't.

RAID will (within limits) help you with recovering from hardware failure, assuming of course that you catch the problem in time and address it promptly. As you quite rightly point out, RAID is not a backup. Also note that all RAID levels are not created equal; RAID 0 provides no redundancy, and can quite easily lower your resilience to hardware failure as all disks in a RAID 0 array must be functional in order for the array to be functional. Lose one, and you lose it all. RAID 5 provides one disk's worth of redundancy, but with today's large disks, the probability of a second failure during the (strenous) resilvering process after a failure is not insignificant.

A "home NAS" is not inherently more reliable than a plain PC built using the same components (and likely is effectively a PC built using very similar components; if you are lucky, with the addition of ECC RAM). It can be made more reliable, but so can your regular PC. (With the exception of perhaps a laptop, for reasons of physical space and hardware compatibility limitations.) For example, my PC has UPS, ECC RAM, and redundant ZFS for storage, all of which contribute to greatly increase storage reliability and all of which are achievable on ordinary hardware. Automated file system level snapshots (ZFS makes this cheap) and automated regular backups provide me with options for disaster recovery.

A single copy of anything, including for example two copies on mirrored disks, is susceptible to being lost to any number of problems. A few examples are (some of these are borrowed from the above linked Server Fault answer):

• File corruption
• Human error (deleting files by mistake)
• Catastrophic damage (someone dumps water onto the server)
• Virus
• Software bugs that wipe out data
• Hardware problems or firmware bugs that wipe out data

If you care about your data, you need at least two distinct copies (including your main working copy); preferably, if you can wing it, three copies (one that you work with, and two for safekeeping). One of these copies should be in a different physical location and ideally one should be on a different type of media than the others. The "distinct copies" part rules out RAIDing the copies; the idea is that a single problem, no matter what kind, must not be able to take out all copies. For a private individual the 3/2/1 scheme (three copies, two backups, one off-site/one on alternate media) is not always practically achievable, but even "just" an external, USB- or eSATA-connected hard disk, hooked to the storage box and regularly updated along with some form of revision history, will go a really long way.

Solution 2

The Synology products all have external ports (USB and/or eSATA depending on model). For small offices we've been happy with back-ups and images of machines going to the NAS[en] and then cloning from the NAS to an external drive connected to it (the Synology software supports this). We then rotate the external drives off-site (sometimes to a home, sometimes to a bank safe deposit).

So the primary data lives on the machines (servers usually), 1st tier backup lives on NAS (a variety of version numbers depending on the kind of data), 2nd tier backup lives on eSATA externals, 3rd tier backup lives off-site.

Comments

[Samuel]

It is a RAID 1 (I edited my post). Periodically hooking a disk to it did not work for me in the first place so I moved to a hopefully more reliable procedure by using the NAS, but I do understand the points you pointed out.

[user]

@Samuel The problem is the "single copy". No single copy is ever safe. "Trivial" RAID 1 is effectively a single copy for anything but complete HDD failure, either of a single sector or the entire drive. (ZFS does better with its total on-disk checksumming, which makes it possible to be certain that a read is valid.) Consider e.g. research.cs.wisc.edu/adsl/Publications/corruption-fast08.htm‌​l as well.

[Samuel]

I would not rely on blindly copying as I've already lost data due to a backup having byte errors and would resort to some checksumming anyway. Even if that would be plain MD5 manually after copying

[Samuel]

I actually tend to use the versioned backup software from my NAS on which I of course cannot execute any checksumming on file level.

[user]

@Samuel If you're seeing unreported end-to-end errors (what is read back is not what was originally stored, and no error is reported by the storage components), then you need something more than a plain RAID mirror, because plain mirroring without checksumming won't be able to tell which copy is correct. (You can recover from that if there's a single error by picking either side of the mirror, but you can't do much if both sides have different errors.) And you should look closely at why those errors crop up in the first place.

[martineau]

IMO, if you really care about your data, you need to have at least two distinct backups of it (besides your main working copy). The more "distinct", media-wise and location-wise, the better.

[user]

@martineau "preferably, if you can wing it, three copies"

[martineau]

Yes, I saw that. It's just that the first part of what you wrote, "two distinct copies (including your main working copy)", translates into only "your main working copy and a backup" -- and why I said "if you really care".

[user]

@martineau Notice how, just before that, it says "at least" :)

[Samuel]

I am actually wondering whether when using a filesystem like ext4 etc on my NAS and copying with robocopy, that additional means are required to ensure file integrity after robocopy and periodically later. AFAIK my Synology has no support for ZFS etc.

[ericx]

The Synologys are Linux. And you can ssh into them and get a shell. But it is a little meager once you're in there. I don't actually know what FS they use, but I'd be surprised if it was ZFS. If you really want ZFS, consider FreeNAS.