ZFS, Actually Storage you can finally reason about.

The ECC Question

concept
depth 2/3 − deeper + surface

First, the term: ECC memory is a pricier kind of computer memory, common in servers, that catches its own rare internal errors; ordinary computers use memory that doesn't. The claim you'll meet arrives with total confidence: running ZFS on ordinary memory is reckless—its self-repair process, fed by faulty memory, will "fix" your good data into garbage. The claim has the shape of expert knowledge: specific, counterintuitive, scary. Actually, here's Matt Ahrens, co-creator of ZFS: "There's nothing special about ZFS that requires/encourages the use of ECC RAM more so than any other filesystem." ECC memory is good engineering—for every system equally. ZFS on an ordinary computer is still safer than an ordinary filesystem on that same computer, because everything ZFS checks is still being checked.

Why the "scrub of death" fails on inspection: for a scrub to overwrite good data, the repair path has to accept a corrupted reconstruction—and acceptance requires the reconstruction to match its checksum. Random RAM corruption producing data that happens to hash correctly is a hash collision by accident—odds so remote they aren't a scenario, they're a rounding error (the jrs-s.net analysis walks the arithmetic). Bad RAM makes a scrub fail to repair things, loudly; it doesn't convert the repair machinery into a corruption engine. The myth requires the safety system to be the weapon, and the safety system checks its own work.