FOB

The ext3 filesystem ate one of my backup drives this weekend, for about the fourth time. To be more exact, the ext3 filesystem on my backup drive got corrupted at least a little bit, and then fsck.ext3 ate the filesystem. So I thought about how I wanted to proceed, and came to the conclusion that it was time to try a new filesystem…

(Does anybody know an authoritative source for the story I once heard/saw that "fsck" is a backronym / rename of the original, more colorfully named BSD filesystem checking utility, done when somebody tried to ship it? In any case…)

The behavior of fsck.ext3 is infuriating, in much the same way Macsyma's is. It asks a lot of yes/no questions, to which there is no conceivable way even the average filesystem guru can really know the answers based on the information given by the program. If you answer no, you're back where you started. If you answer yes, things are probably going to get worse from there on out. I'm not at all confident the fsck.ext3 heuristics are any good anymore. I certainly can't remember the last time fsck.ext3 ever recovered a corrupted volume rather than trashing it.

Bah.

Anyway, I started digging around for a filesystem that is:

• Well supported on Linux.
• Maximally reliable / recoverable in the face of disk corruption, even at the expense of performance.
• Completely compatible with Linux filesystem semantics.

ReiserFS, Reiser4, XFS and ZFS seemed like the obvious candidates. However, all four seemed to have quite a few reports of serious trouble with reliabilty, support, or both.

Then JFS popped up. It apparently was originally designed by IBM for AIX, with exactly the design goals I wanted. A quick Google search turned up no obvious, recent, or alarming flaws. The fact that IBM is supporting it felt good to me, since I have friends inside LTC who presumably can help me.

One of my backup drives is now running JFS. Making and fsck-ing the filesystem was dead easy, as expected. If anyone knows any reason why I should flee, I'd love to hear it. Otherwise, I'm going to give it a try.

One nice thing about trying this on a backup drive is that if it goes wrong, nothing very bad will happen. Further, these drives get abused quite a bit, which is presumably what was corrupting ext3 in the first place, so it will be a reasonable test of reliability. Performance is a total nonissue here, since I'm running the drive over a USB 2.0 HS connection which is totally the bottleneck anyhow.

I'm happy at the moment, but others' thoughts would be greatly appreciated.