James Wiebe, of WiebeTech, has written a terrific white paper that exhaustively analyzes the reliability of external enclosures, detailing the factors that contribute to failures.
As we’ve known for some time here at Shirt Pocket, large capacity external drives that use multiple physical drives in a single enclosure to boost their performance and capacity are much less reliable than single drive units.
As far as I’m concerned, anyone who relies upon hard drives for storage, backup or otherwise, should read and understand this paper.
Thanks for writing it, James!
09 May 2006 at 11:01 am | #
Can you provide a few recommendations for external enclosures?
09 May 2006 at 11:05 am | #
In general, we recommend LaCie d2/d3 (non-multi-drive), OWC, WiebeTech and Maxtor OneTouch II/III FireWire or FireWire/USB drives. (Note that there are models of the III that have multiple drives—make sure you get the III *m*, which allows the extra drive to be used as a mirror, not the III *s*, which only allows striping. Also note that we do support the OneTouch button on the Maxtor drives: see the SuperDuper! FAQ (Help > Frequently Asked Questions) for the technique.)
09 May 2006 at 11:13 am | #
Thanks for the quick follow up!
09 May 2006 at 11:19 am | #
One more question. Can you recommend just an external case(no drive, I have one already)?
09 May 2006 at 01:25 pm | #
Both WiebeTech and OWC sell cases or drives and cases.
09 May 2006 at 08:38 pm | #
Now, hold on. Multiple drive enclosures by definition have more points of failure and will be more failure prone. However, if we’re talking multi-drive enclosures in a JBOD, there is no real higher probability of *any single drive* failing than in a single-drive enclosure. Put another way, having four one-drive enclosures does NOT make it more reliable than a single four-drive enclosure. In fact, it may make it less reliable as you’ll likely have more bridgeboards involved.
What I was hoping would be covered was discussion of the construction on various enclosures, especially with regards to heat dissipation. This is an area where enclosures vary wildly, and will also quickly fry a drive if used heavily (backup, video, etc). I have two identical enclosures that fried their drives (quite literally) due to lack of ventilation. Now I have a four-drive enclosure that has fans out the wazoo, and I can use them non-stop for hours without them getting so much as warm. The enclosure design can make an enormous impact on reliability, and that was regretably not covered.
One thing also not emphasized here is hardware versus software RAID. Software RAID is performed by your OS, and uses CPU power to do its work. So, this means a software RAID set will only be usable by the same OS, and will have a slight performance impact. For this reason, you rarely see software RAID-5, as that would have a severe performance impact (running parity calculations on each bit is expensive!). Hardware RAID makes the whole thing invisible to the host - it sees a single drive (however you set it up), and it takes care of any calculations or overhead. However, you rarely see hardware RAID outside of simple 2-drive RAID-1 enclosures, or high-end server enclosures/adapters.
As for RAID levels, let me add some additional perspective. I work in enterprise hosting, and run personal servers with SCSI RAID at home. RAID-0 (striping/spanning) should only be used for data that is either temporary or backed up well. It increases the points of failure, but does nothing to protect against such failure. RAID-5, meanwhile, has a good balance of reliability and storage use (you can use the capacity of N-1 disks), but it carries a huge performance hit, especially in writing and duplicating. On a 6-disk SCSI system that was blistering fast on RAID-0+1 (three disks striped in RAID-0, then mirrored to three other disks in RAID-1), reconfiguring as RAID-5 dropped file duplicate rates to less than 1MB/second. Yes, that’s slower than your home network. Much slower. I agree with the artile that RAID-1 truly is the best if you can afford it. It cuts your capacity in half, but completely protects against a physical drive failure.
Note that I said physical drive failure - as was mentioned early on in the paper (and then not revisited), if the OS or your software has a bug and wipes the directory or otherwise mangles the bits, RAID will not help you, as it will dutifully copy that info to the other disks. The lesson here is you still need backups (in fact, you should always have a comprehensive backup plan before even thinking about RAID - especially so for RAID-0). So use SuperDuper already!
02 Jun 2006 at 03:37 pm | #
The paper is not terrific. Wiebe’s math is extremely bad. By his equations, a hypothetical enclosure with 100 hard drives has a probability of failure over 200%—well any result over 100% proves the equations wrong. (In reality, with his numbers that 100-drive enclosure should still have less than 90% failure probability.)
02 Jun 2006 at 05:40 pm | #
An excellent point, Guillaume: you should drop him a note so he can correct his paper…
25 Sep 2006 at 01:12 pm | #
Yeah, reads like an ad. I am becoming less and less impressed with Wiebe as time goes on; they seem to be going down the Granite Digital road of selling essentially commodity parts with the addition of decent support at highly inflated prices.