(Backblaze engineer here.) Online backup is a unique application for hard drives. This is not a high transaction environment like a database getting pummeled with SQL queries. In online backup, we write the encrypted data only once, remember a SHA-1 to verify the data later, then we read it once a week to make sure the data still has perfect integrity (the SHA-1 must match). We now have over 10,000 hard drives spinning in our datacenter, some for over five years, and we've collected some pretty detailed statistics on drive failure rates vs many parameters. For our application (online backup) drives from external enclosures and consumer drives last every bit as long in our environment as "enterprise" drives.
I'm sure you guys keep track of read/write statistics as well; have you thought about spinning drives down on full BB pods until a restore is requested (similar to how Amazon Glacier's storage system works)?
I read once that the majority of drive failure happens during transitions to or from a running state, and thus it is counter-intuitively safer to just leave them running all the time.
I am sure there is some mathematical formula that will express the risk based on how infrequently one expects to spin the drive up (and then down again), but I don't know it.
Backblaze is hosted out of a SF datacenter, where power is much more expensive than other US locations; I assume someone has done the math whether its cheaper to replace failed drives from frequent spinups/downs vs burning the power to keep them spinning.
