ZFS, Actually Storage you can finally reason about.

Deduplication

concept
depth 2/3 − deeper + surface

Deduplication means storing each identical piece of data only once, however many times it appears. The reasoning for turning it on feels airtight: ZFS already fingerprints everything it writes, so noticing duplicates should be nearly free. Actually, noticing isn't the cost— remembering is. Dedup requires a directory of every piece ever written, consulted on every single write, which means kept in fast memory: a few hundred bytes per entry ("slightly more than 320 bytes of memory"—OpenZFS docs), forever. When that directory outgrows memory, every write has to go ask the slow disks first—the storage equivalent of walking with your shoelaces tied together.

Run the arithmetic before the feature: a few TB of unique 128K blocks is already multiple GB of dedup table that must stay resident—RAM that was being your ARC. In almost every homelab, the RAM dedup consumes would buy more value as cache than dedup saves as space, and plain lz4 compression already captures much of the same win (similar blocks compress well) for approximately zero cost.

Where it genuinely pays: high-duplication, high-value-per-TB data—dozens of near-identical VM golden images, certain backup corpora—on machines specced for it. If you're not sure you're that case, you aren't. Estimate first: zdb -S tank simulates the dedup ratio without committing to anything. Under ~2×, close the tab.