Comment by hedora
9 days ago
The operational expenses of this stuff dwarfs the hardware cost. For the tape mountain, you need robots to confirm the tapes still work (mean time to detection of device failure and recovery are key for RAID durability computations). So, someone needs to constantly repair the robots or whatever.
If I was being paid to manage that data set, I’d probably find two enterprise storage vendors, and stick two copies of the data set on them, each with primary secondary backup. Enterprise flash has been under a dollar a gigabyte for over a decade, so that’s under $1.7M per copy, amortized over five years. That’s $700K per year, and one of the four copies (at 3-4 sites) could be the primary store.
(I can’t be bothered to look up current prices, but moore’s law says there have been six capacity doublings since then, and it still applies to flash and networking, so divide my estimate by 2^6 — so, ten-ish grand per year, with zero full time babysitters required).
even with dual vendors, you'd have to still put in place a backup/restore procedures (with the associated software, which may need to be custom). Then you'd need regular testing. These operational concerns will basically double the cost yearly, probably.
You'll need permanent staff to oversee this, too. This will add at another ~500k+ to your annual expenditure.