This is your best bet: Distributing blocks to all servers evenly. Using RS to encode your blocks to limit storage space requirements, and make recovery obvious.

This is your best bet: Distributing blocks to all servers evenly. Using RS to encode your blocks to limit storage space requirements, and make recovery obvious. Also, make sure you don't buy all of your hard drives from the same end manufacturer. Most of these types of systems assume random hard drive failure... turns out drives can have highly correlated failures.