I worked for IX systems for a few weeks (primary developers of freeness and truenas), they had some clients who had High availability arrays of 2PB to 6PB per High availability head unit. (In under 2 seconds could recover from a motherboard failure with no dropped connections, continuing existing data transfers).
Here is a write up I made for how to set up a large disk array for maintaining production performance while resilvering under during multiple disk failures.
There is also discussion about how to setup the arrays with hot spares and vdevs spanning enclosures to recover from someone dropping a disk shelf down a stairwell.
dedupe:
BTW at this point SATA and SAS SSDs were starting to reach the 256GB size, and HDDs were 14TB. But much of the data is still useful.