[stoppingdown]

WIth Amazon Glacier costs are in the magnitude of $0.01 per gigabyte per month. You don't have a real-time access to data and thus the backup process must be managed with some more care, but being a long-term backup this is hardly a problem.

 

The only reason for which I'm not using it yet is about privacy - I don't sell my photos, there are no photos which hold secrets or privacy stuff, still I don't like to accept the fact that I put my stuff on the web and some people have more or less unauthorised access to it (this particular scenario, anyway, should be manageable with an encryption software run locally, before putting the stuff on the cloud).

 

I recommend using multiple approaches. The backup process for my photos is the same I use for all my data: everything is on my laptop (with two disks), it's cloned bit-per-bit to another laptop (the older one), I have an USB external drive that gets incremental backups several times per day; then everything is cloned to a RAID 6 storage (5 disks, 2 can fail with no loss of data); after some time, when the decimation of photos has been completed, they are archived to both a set of DVD (4GB per disk), DVD hi density (8GB per disk), twice. The two sets are kept in two different places of the house. I used to keep one set in another house in a different town where I frequently worked, but things have changed in the past two years. Still, I feel decently covered by this strategy. My house burning, or an earthquake, would destroy everything, but I have a practice to keep always at hand a small bag with all the extremely important stuff (such as the wallet, with documents inside): it also contains the USB backup unit. 

 

ZFS has been mentioned, and it's a good point. Some years ago it seemed that Apple was going to include it in Mac OS X, but they didn't. Unfortunately HFS+ is very unreliable, in the past I've seen at least twice silent data corruption happening. People should be aware of this in case of incremental backup to a rewritable media: the risk is to have one file corrupted to be inadvertently backed up and overwriting the good backup file. That's why I have both, backups on rewritable and non rewritable media. Furthermore, I take advantage of the fact that RAW files never change, so I always compute their fingerprint (e.g. MD5) and periodically compare them. When a MD5 changes, something is wrong - stop, think, find the good file and restore. This practice should be also periodically used to check the integrity of optical media, because they last long but aren't eternal.

 

PS Yes, a RAID controller can fail, but I don't see the problem: we're talking about stuff which is off-the-shelf, so buying a replacement is not a problem. Just keep an eye on the market to check whether the product goes end-of-life and spare parts are no more sold. Also, if you use RAID units with a documented protocol, ever in the worst case you'll find a professional service capable to recover the data - it would cost you a few bucks, but this is an extreme case.