Rick Owens

@mhoye@cosocial.ca FWLIW, as a practical example, a few years back I used an old Lacie 1U enclosure to make a boot drive for a FreeBSD BackupPC server. Inside the enclosure I put a power strip, a USB hub, and 4 USB enclosures for 2.5" drives. I put boot code on each drive, then made a 4-way mirror using a ZFS partition on each drive. As expected, several of the drives failed over time, but it was easy to replace a dead drive with a 'new' one - just partition the new drive, put the boot code on, and replace the device in the zpool.

@mhoye@cosocial.ca Just to clarify, you can also use systems that do RAID-like checksums rather than mirroring, trading computation time for space. You can do that at the block level, you could even do it with ZFS by partitioning the disk n ways, then making a RAIDZ3 zvol from those partitions. But there's little point, because again it focuses on only one fault type and ignores others that are just as likely if not more so.

@mhoye@cosocial.ca How hard do you want to go/how many nines are you willing to pay for? On a single system, ZFS will give you pretty good durability; use enough drives and HBAs, esp. different types of drives and connections, and the likely failure points become the CPU, power supply, etc. (Seriously, you want to do a 10-way mirror with spare drives hot and ready to go, ZFS can do that; it will let you go to absurd levels and then keep right on going, assuming you have the budget for it.) If you want more nines, you'll need to go bigger - HA networking with multiple computers, multiple switches, geographic separation, etc., etc. How far down the rabbit hole do you want to go?

@mhoye@cosocial.ca @petko@social.petko.me @mms@mastodon.bsd.cafe Are you looking at it from general curiosity or a particular use case? And are you factoring in backups of any sort?

@mhoye@cosocial.ca Hmmm.... ZFS (and I believe other systems) can do multiple copies on disk (literally via copies=n for n greater than 1). If your only worry is about failing sectors, yay, you're done. 😦 But... that's a bad idea. Yes, sectors can fail, and yes, having multiple sector copies with ZFS means you'll read good data as long as at least one good copy exists. But in the real world, are you more worried about failing sectors or failing drives? Or more generally, what's your exact use case and threat model? Focusing too tightly is more likely to bite you in... uncomfortable places... than to make you happy in the long run.

@gemlog@friendface.kalum.ca @mhoye@cosocial.ca FWIW, I'll disagree, sort of, if you'll forgive me being pedantic. "My question is only, is there a filesystem out there that lets me sacrifice X drive space for Y error correction capabilities." Technically the answer is a yes, ZFS and others can give you all the protection you want against single sector errors. copies=100 and/or extreme parity layouts can give you your data even if you can't read 99% of the sectors. But, there's always a but. The problem is statistics, or 'Mother Nature always sides with the hidden flaw.' if you prefer. You asked about extreme methods of warding against one failure type, but as gemlog said, we respond to the implied (or perhaps that should be 'common') question about best overall data protection methods.

@gemlog@friendface.kalum.ca @mhoye@cosocial.ca Hey, I resemble that remark!