Your SATA RAID has a 56% chance of data loss

http://alumnit.ca/~apenwarr/log/?m=200809#08

by oxyona • 18 years ago

55 25 18 years ago

ZFS however would protect against this, in that it checksums each chunk of data. And "zfs scrub poolname" will go through and do the checksums for you in the event you are storing a lot of data and not necessarily reading/writing it often.

spydez• 18 years ago

Well, maybe I should look into transition my home server from Ubuntu and XFS to Nexenta or OpenSolaris and ZFS...

I'd hate to lose some of the data on that thing, though I'm currently not doing any RAID. Just a disk of important stuff rsnapshotted every 4 hours to another disk.

lallysingh2• 18 years ago

I've been running opensol/zfs for almost a year now. Zero problems, good perf & reliability. My research is on two mirrored sata/160gb drives. The most important stuff is on a FS on there with copies=2, making 2 copies on each drive.

Also, backing up zfs is really, really easy. 48 lines of shell scripts do daily incrementals and weekly snapshot backups.

But, do yourself a favor and put /usr/gnu/bin first in your path. oh, and alias tar=gtar. Old solarisisms will otherwise be irritating.

attack3• 18 years ago

No, but it will notify you of a problem. However, I suspect that smartmontools would do an even better job of notifying you of a problem before it even occurs.

If you want something that will actually protect against bit level errors well, generate some extra data using an appropriate Error Correcting Code for your important archives/files. Something like what par2repair generates. This is incredibly slow compared to simple erasure codes like RAID uses, though. There are some filesystems that do this for you automatically, I wrote one:) Actually, other than that I don't know of any filesystems that do this.

If the above is impractical for performance reasons, Google recommends mirroring everything on three disks...

iigs• 18 years ago

As theoretically awesome as ZFS is, the math behind the human factors are more important than the math behind the computer failure rates, and this is something I don't believe we can fix with our current development methodologies.

It looks like Joyent was able to rebuild most (all?) of their data in their ZFS data loss situation, but it could have played out drastically differently.

http://discuss.joyent.com/viewtopic.php?id=19430

lallysingh2• 18 years ago

Note that it mentions that SATA has a higher bit-error rate than SAS/SCSI. Apparently through the use of fewer ECC bits.

But yeah, ZFS rocks.

ars2• 18 years ago

You could also set smartmon tools to do a short self test every day, and a long one (which reads the entire disk) every week.

At least you'll know if there was an error.

attack3• 18 years ago

This is what keeps encryption impractical, especially for laptops.

lonestar• 18 years ago

Why? Assuming you're using CBC mode within each sector, you wouldn't lose any additional data due to a sector error compared to an unencrypted disk.

attack3• 18 years ago

These are bit level errors. So instead of losing a bit or two, you lose a massive chunk. That's the difference between a strange character in your Doc file and losing the whole thing.

I think it's 128 bits for truecrypt, although internet searches point to many people using higher (1024 and 4096). And if the bit level error rates are as high as he says, looks bad.

cstejerean• 18 years ago

http://en.wikipedia.org/wiki/Disk_encryption_theory

ars2• 18 years ago

Wow. If you have an 8TB raid, that's 7TB of data so 61,572,651,155,456 bits

A hard disk has a bit error rate of 10^15:

1,000,000,000,000,000

That's about 16 times higher.

That's horrible.

Taken to a single 750GB disk that means you can read the entire disk an average of 155 times before you'll encounter an error! That's shockingly bad.

Andys• 18 years ago

I am, among other things, a sysadmin who babysits several RAIDs, including a 10TB one.

I can confirm that the large capacity SATA disks have a noticeably high error rate, which I have never seen before. It used to be that RAID protected you from disk failures, but we aren't seeing that nearly as much now as we are unrecoverable bit errors on otherwise healthy disks.

I have set all our RAID controllers to scrub the data and compare the mirrors every weekend. In the past couple of years we've only had a couple of unrecoverable read errors, but many recoverable read errors.

Recoverable read errors are still bad, the disk gets taken offline and out of the RAID set for up to a minute while the controller retries many times until the data can be read. Then the sector is remapped to another part of the disk.

So yes, it is a real problem, but it isn't too hard to manage in this era of filesystem snapshotting plus high speed RAID6 controllers. In return we've been given huge amounts of storage space.

JoelSutherland3• 18 years ago

I keep waiting for a product that solves the failing hard drive problem at the consumer level. For most people I know, the most likely cause of data loss is hardware failure.

Are any manufacturers offering a package that contains multiple hard drives but behaves as one?

If performance isn't a factor, is there something I don't understand that makes this impossible?

wmf3• 18 years ago

Drobo? Time Capsule?

JoelSutherland3• 18 years ago

Drobo is along the lines of what I am looking for. (I use a D-Link DNS 323)

Instead though, I want it to have a SATA interface so the problem would be solved in hardware. Time Capsule fails if the disk fails (and backups aren't stored elsewhere).

briansmith2• 18 years ago

* Most hard disks are actually multiple hard disk platters packaged together. Even with today's systems, the filesystem can multiplex data across the platters so that the platters mirror each other. Even on a single-platter drive, the file system could just mirror the data on the same platter.

* People want increasingly-higher-capacity, smaller, cheaper, thinner, lighter, and cooler laptops. Adding a second magnetic hard drive into a laptop runs counter to all those goals, as does the kind of mirroring described above. The consumer market demands higher capacity over safety.

* Solid state hard drives seem to be the future, at least for laptops. It might be practical to install dual SSDs into a laptop since they are smaller and lighter. It might even be practical to have a rotating set of three as a backup system. Eject hard disk #1, then insert hard disk #3. Hard disk #1 becomes the backup. Hard disk #3 gets synced with hard disk #2 until they mirror each other. Then, relabel the disks 1->2, 2->3, 3->1.

* People are being encouraged to save (or at least back up) their data on the network if it is really important.

JoelSutherland3• 18 years ago

"The consumer market demands higher capacity over safety."

My contention is that this is beginning to change. Short of video (and at the risk of being wrong in 5 years) there is nothing to do with a 500GB drive.

I think given the trade-off, consumers would choose safety.

thorax• 18 years ago

Windows Home Server actually uses a duplication scheme that makes sure all your machine backups (and shared folders) are stored, in full, on more than one drive on the server. It behaves pretty much like one big drive and duplicates across all the drives as needed.

It's pretty likely a consumer won't need more than this since that equates to three copies of your data at any given time (laptop, server disk B, and server disk C).

sabat• 18 years ago

I don't have any hard data, but my BS meter is going off.

keyes• 18 years ago

Any decent hardware RAID controller has a background "patrol read" which takes care of this problem.

aparrie• 18 years ago

Anyone use ZFS on Linux?

http://www.linuxworld.com/news/2007/061807-zfs-on-linux.html

Your SATA RAID has a 56% chance of data loss

Related Stories