Why does a RAID-0 volume randomly fail with healthy drives

Drives, like most components, have a bathtub curve failure rate. They fail a lot in the beginning, have a relatively low failure rate in the middle, and then fail a lot as they reach the end of their life.

Just as the whole drive follows this curve, particular areas of the disk will also follow this curve. You'll see a lot of sector re-allocations in the beginning of using the drive, but this should taper off. When the drive starts to fail at the end of life it'll start losing more and more sectors.

You don't need to worry about 6 (depending on the drive - consult the manufacturer), but you need to watch and see the frequency of each new reallocation. If the deterioration accelerates or stays the same, worry. Otherwise, it should be fine after the initial break-in period.

-Adam

It depends on whether you're talking about server-class gear or desktop-class gear.

If it's a desktop machine built with your own money and off-the-shelf drives, and you're not worried about compatibility, then yes, your strategy is sound. Every X years, go out and buy all-new drives to replace your current drives. They're going to be faster, quieter, and larger. You could replace the drives individually, letting the array rebuild itself, and then when the rebuilds are complete, reconfigure your array to be larger. (Not all raid adapters support operations like this - online rebuilds and size changes.)

If it's a server-class machine like an HP Proliant or IBM System X, it gets more complicated. You may need to use hard drives on the compatibility list for your raid adapter. In that case, the drives are going to be expensive because they're probably no longer produced, or they're just plain expensive to begin with on server-class stuff anyway. Even worse, you might be buying refurb gear from your reseller and not knowing it - this isn't uncommon with server resellers.

Plus, you may be discarding drives with perfectly good lifespans and replacing them with drives that are destined for trouble. Rather than proactively replacing those, it makes more sense to build the server with a hot spare to begin with, and make sure your raid array supports automatic rebuilds using a hot spare. Then the rebuild will happen before you even get out of bed to make it into the datacenter, and you can replace the dead drive at your leisure without spending money or time.