Comment Re:Good luck with that (Score 1) 171
It's not just the OS lying about a write being complete. The hard drive controller can lie and say the write is complete when really just queued up to be written. Same for the hard drives themselves. Most modern hard drives have an internal write cache that's enabled by default.
Normally, it doesn't cause too many problems, but for things like transactionaly consistent databases, it's a really really big deal when the H/W lies about a write being complete.
At work, I actually had to test this using a remote logging disk write test driver. Under high write load, we could pull the plug and the test driver could detect that a write operation was logged to the remote machine as completed, but the write never made it to disks. This could cause a lost transaction log write would corrupt the database about 90% of the time (we used diskchecker.pl to test this. See https://brad.livejournal.com/2... ). We found the obscure disk controller command that turns off the hard drive internal write cache, and it completely fixed the problem.
Note that modern SSDs have this problem too. Some "enterprise quality" SSDs have a capacitor power the SSD for the few seconds it takes to flush it's write cache. (See https://ec.kemet.com/ssd-hold-... )
