as one who just experienced a catastrophic failure of a raid 5 “backup” I would offer that rendundant copies using raid 0’s is probably more reliable. The odds of a second drive failing in a raid 5 while rebuilding one that has failed grows dramatically as drive sizes increase, and in fact most data centers have/are moving to raid 6 for that reason - and predicting that within a few years even raid 6 will be problematic. While theoretical estimates probably overstate the problem, it is a problem. A raid 6 can lose 2 drives, so if another fails during the rebuild you are OK. The main advantage of a raid 5 is the data can still be accessed while it rebuilds and it is cheaper than the same amount of storage setup as a raid 1 or using full duplicate/cloned backups. Other than data centers or large corporate servers, most don’t need that feature, and raid 5 is not a backup solution. Another problem with raid 5 is rebuild times are very long ... we’re talking many days. Lower priced raids take 24-36 hours per terabyte of data to rebuild. (here is
one article discussing these issues with raid 5 )
Additionally, hardware raids while they sound great, also have caveats (an “error” of a single block on a raid 0 on one of the drives can mark the entire drive as bad, resulting in total loss). A software raid seems to handle this better. There is a slight speed hit and possibly a little extra processor overhead, but neither are noticeable. Certainly there are downsides, but after some experience I have opted for software raids. (as an example, see
this article)
I had a 4 drive raid 5, and one drive failed. Fortunately I learned long ago that raid 5’s and their redundancy is not intended as a backup method but as a keep the data available solution, so that raid 5 was cloned to a second raid 5 each night. Anyway, a new drive was inserted and somewhere during the rebuild a second drive failed, because the unit now is stuck in rebuild mode, and it doesn’t appear as the drive will be recoverable.
I already had a LaCie Big 5 unit, so I bought
2 OWC JBOD thunderbolt 2 cabinets along with eight 4 TB HGST(Hitachi) 7200 RPM drives configured each box using OS X soft raid as a set of raid 0’s. (some more research I did seems to show the hitachi/HGST drives are the most reliable,
see this.)I debated on doing a raid 10 so that all changes are reflected to both drives, but decided an issue with that could be a corruption which gets mirrored. I opted instead to set the system up as a daisy chain backup, my main raid 0 is cloned over to the backup #1 at 3:00 am using Carbon Copy Cloner, when that finishes it triggers a clone from backup 1 to backup 2 using carbon copy cloner. So each day I start with 3 identical copies of the same data, and by changing a few symlinks on the SSD startup volume which maps some of my home folders over to the raid, I can start up with any 3 of the drives if necessary . The downside is if my main raid goes down during the day I lose anything I have done that day. Normally that isn’t that much, since I’m working at a second location with clones of various folders, which are then cloned over to the main drive when I get to my Mac Pro workstation.
The OWC thunderbolt 2 JBOD box is about $100 less than the one with the raid hardware built in. I’m much more comfortable with my setup now, although I may move the 3rd backup to another site, and try to setup my own internet backup scheme to keep it current. Haven’t decided if that is practical or not yet. (also looking at various cloud/online options to backup just some critical folders). I’m also planning on setting up a backup to an SSD of all my critical data which can be stored at a second location. I believe SSD’s which see very little data change offer the best long term backup solution, since they have no moving parts. And as SSD’s get cheaper I may look at creating a SSD raid 0 volume which can be a main work volume (which of course would be cloned to my main raid 0, and so on ). But right now I’m seeing 600MB/sec throughput on my 4 disk raid 0 ... I’m not sure I would see any really improvement in LR or PS with this.
So this may be beyond the budget of most, but the basic idea to me is raid 0’s when you need speed, redundant copies rather than parity raids (5 or 6) is probably more reliable and less troublesome if the data doesn’t have to stay live 24/7.