Comments on: Hammer and mirrored disks

By: Justin Sherrill

Justin Sherrill — Sun, 03 Aug 2014 15:49:21 +0000

Yeah, you’re right – the implementation matters. As I said, it’s not reasonable – just my reflex. I worry about conflicts between what the hardware thinks is there and what the software thinks is there; the readability of the remaining data if there’s a dead disk, etc.

I suspect part of the popularity of cloud services is because people can blithely give up on caring about redundancy because the problem is shoved on someone else.

By: Rahul Siddharthan

Rahul Siddharthan — Sun, 03 Aug 2014 13:05:26 +0000

“software RAID means that one scrambled bit of data on one disk means that the whole group is messed up”
Doesn’t that depend on the implementation? My impression was that it shouldn’t happen with a well-implemented RAID 5 or 6 (distributed parity). And in particular, with something like ZFS that does aggressive checksumming, it shouldn’t be a concern.

By: Justin Sherrill

Justin Sherrill — Thu, 31 Jul 2014 12:36:25 +0000

Oddly, I always had the opposite impression – software RAID means that one scrambled bit of data on one disk means that the whole group is messed up, and since more disks means lower MTBF overall, you are more likely to lose everything.

That’s not necessarily a reasonable worry; just the impression I’ve had. In any case, proper backups (which is one place Hammer mirroring can come in very handy) is the best thing to have no matter your disk setup.

By: Lazarus

Lazarus — Thu, 31 Jul 2014 11:00:29 +0000

I’ve always been told that hardware raid is evil. Say your controller dies, you need to get the exact same hardware in order to recover data on the raid volume. With software raid, you just plug the disks into another machine. Moot point i guess since you’d likely also have proper backups.