DaPlumber
Patron
- Joined
- May 21, 2014
- Messages
- 246
I want to back up the message cyberjock et al have been proselytizing: ECC-RAM is required for ZFS systems.
I've been messing with ZFS since before we got compression, so around zpool version 5 and snv_55 IIRC. I was privileged to be working at Sun at the time and a lot of the internal literature stressed that memory that could detect errors was an architectural design assumption for ZFS. Obviously correcting single bit errors rather than halting the system is preferable, hence ECC. This isn't exactly a radical requirement in the server game anyway, my favorite example is how would you feel if your bank couldn't detect a flipped bit or failing cell in, say, your account balance? To contradict a vicious rumor: It's not true that there isn't a tool for recovery, but using the zfs debugger is a skill on the level of live patching assembly code with a monitor. I.e. there's a very short list of people who can do it, you're not on it, and they have better things to do with their time. It's easy to see why that rumor got started, because for all practical purposes it's true! :D
That said people use ZFS with non-ECC RAM all the time and on things like laptops without even a mirror. That's OK as long as they understand the tradeoff is that it cannot be viewed as a safe repository. Critical data (or the whole system) needs to be backed up elsewhere, and with good incrementals or snapshots. This should be the default approach with things like laptops anyway as there are so many other things that can go wrong, get lost/stolen etc.
tl;dr Take it from someone who knows ZFS: Listen to cyberjock et al and build your production systems with ECC-RAM. Without ECC it's a question of "when" you're going to lose a pool/s, the whole pool/s, not "if".
I've been messing with ZFS since before we got compression, so around zpool version 5 and snv_55 IIRC. I was privileged to be working at Sun at the time and a lot of the internal literature stressed that memory that could detect errors was an architectural design assumption for ZFS. Obviously correcting single bit errors rather than halting the system is preferable, hence ECC. This isn't exactly a radical requirement in the server game anyway, my favorite example is how would you feel if your bank couldn't detect a flipped bit or failing cell in, say, your account balance? To contradict a vicious rumor: It's not true that there isn't a tool for recovery, but using the zfs debugger is a skill on the level of live patching assembly code with a monitor. I.e. there's a very short list of people who can do it, you're not on it, and they have better things to do with their time. It's easy to see why that rumor got started, because for all practical purposes it's true! :D
That said people use ZFS with non-ECC RAM all the time and on things like laptops without even a mirror. That's OK as long as they understand the tradeoff is that it cannot be viewed as a safe repository. Critical data (or the whole system) needs to be backed up elsewhere, and with good incrementals or snapshots. This should be the default approach with things like laptops anyway as there are so many other things that can go wrong, get lost/stolen etc.
tl;dr Take it from someone who knows ZFS: Listen to cyberjock et al and build your production systems with ECC-RAM. Without ECC it's a question of "when" you're going to lose a pool/s, the whole pool/s, not "if".
