There Might Be Another EXT4 Corruption Bug

[makomk]

As far as I know, ZFS is the only usable alternative available that provides solid data integrity guarentees.

Yep, it provides an 100% solid guarantee that it'll detect data integrity issues. It doesn't guarantee that it won't corrupt your data, or that you'll be able to get a single byte of your data back after even the most trivial metadata corruption, and it doesn't even guarantee that it'll bail out nicely rather than kernel-panicing the moment you try and mount a corrupt FS (as I believe various users of ZFS on Solaris have found to their costs). It's guaranteed not to let you continue using your FS once it becomes corrupt, though.

[ryao]

Thanks, i will try it in my nest build....i might need to build a Linux box around a mini-ITX MB sooner than i thought and will use ZFS on it...

Is ZFS by any reason slower than EXT4 ?

Data integrity is a must but i would like to know if there is a speed penalty if i also use it in a gaming rig....



TIA,
AJSB

ZFS will likely have a small penalty in write performance. Read performance should be dramatically better. Your games should be almost entirely loaded in memory, so filesystem performance should not matter.

https://forum.sabayon.org/viewtopic.php?f=86&t=28059&p=154204#p154165

Yep, it provides an 100% solid guarantee that it'll detect data integrity issues. It doesn't guarantee that it won't corrupt your data, or that you'll be able to get a single byte of your data back after even the most trivial metadata corruption, and it doesn't even guarantee that it'll bail out nicely rather than kernel-panicing the moment you try and mount a corrupt FS (as I believe various users of ZFS on Solaris have found to their costs). It's guaranteed not to let you continue using your FS once it becomes corrupt, though.

Your statements are consistent with an attempt to fabricate criticism by someone whose sole experience with filesystems are those in the same category as ext4.

First, it is impossible for "trivial metadata corruption" to kill ZFS because ZFS writes all metadata at least twice and does its best to space the redundant copies far apart. Second, the concept of mounting "a corrupt FS" makes no sense with ZFS because of how it is organized (pool import and mounting datasets).

[AJSB]

ZFS will likely have a small penalty in write performance. Read performance should be dramatically better. Your games should be almost entirely loaded in memory, so filesystem performance should not matter.

Thanks...i was concerned more with read speeds than write speeds...

[Candide]

XFS

I first encountered the EXT4 bug about six weeks ago. It's actually quite severe, crashing two of my systems due to file system corruption. At first I wasn't sure what it was and thought my hard disk might be going bad, but after it happened on two separate machines, I eventually figured out that EXT4 was the culprit by trial and error (I restored one of the machines to XFS, it's previously used file system, and the crashing went away). I have found XFS to be solid and it's very fast. If you want to format a hard drive with XFS, you need the package "xfsprogs" to be installed before running the installer. DistroWatch is going to run a "how-to" article this coming Monday.

I don't recommend EXT3. It's seriously slow, and lacks a lot of features. It was a first best attempt to add journaling to EXT2 and was never meant to be more than a temporary file system until EXT4 was released. And EXT4 was very good, until this bug(s) crept in.

Some have mentioned ZFS, and everything I've read about it sounds good. The only trouble is that most distros don't have support for it compiled into their kernels. It does exist, but you'll have to hunt around for a distro that has it.

[AJSB]

Thanks for the info...no matter XFS is "old" , i never tested it

Indeed, ZFS doesn't seem to be supported at install level by SLACKWARE but XFS seems so...i will try XFS because of that and also because of other things you said.

Thanks also for the heads up about the article in Distrowatch about it....looking forward with interest for it .

[AJSB]

Nice XFS review at DW

...and one more reason to stay with Slackware....xfsprogs is in the DVD by default

[Pallidus]

Hey guy who owns phoronix:

I red a blog post from a chinese kernel dev and he was saying you were just trolling for adhits and should stfu...


them seem like fighting words to me

[mvidberg]

I've been having issues with ext4 data corruption in the 3.5 kernel (in the latest Ubuntu) as outlined in https://bugs.launchpad.net/ubuntu/+s...x/+bug/1063354

Not sure yet whether this is a Ubuntu only issue or just my combination of hardware... but I've been running on 3.4 kernel now with no more indication of data corruption in my syslog file.

[kernelOfTruth]

I've been having issues with ext4 data corruption in the 3.5 kernel (in the latest Ubuntu) as outlined in https://bugs.launchpad.net/ubuntu/+s...x/+bug/1063354

Not sure yet whether this is a Ubuntu only issue or just my combination of hardware... but I've been running on 3.4 kernel now with no more indication of data corruption in my syslog file.

looks more like a hardware issue (harddrive, sata-cable, sata-controller, memory, etc.) to me than the filesystem per se


if you've the time & backups

try using btrfs or xfs and see whether it occurs also with them - then you'll know for sure if your data is safe with the device you're using or it's a ticking time bomb ...

[gamerk2]

In Window$ we download patches for what we even haven't a clue for what they are.....

You can generally check the MSDN page for more info if you so desire.

and yes, one of the several reasons i dropped Window$ was data integrity because several times some games locked my rig and only way to switch off was to hard reset making data corruption....

Based on past experience, this typically indicates an unstable memory subsystem, usually due to the mobo's NB not being able to handle four DIMMS of high speed RAM. Some mobo chipsets (NVIDIA 680i/780i/790i, and most lower end ECS/Biostar mobos) were notorious for this once people started using all the RAM slots. Easy test for this would be to run Prime95 for a few hours and wait to see if that comes back with a rounding error (indicating memory corruption occured). Ran into this issue with my old crappy XFX 790i Ultra; had to overvolt the NB to 1.3V and underclock the RAM for stability. Ran fine on linux though (or rather: Could not reproduce. Didn't have an equivalent to Prime95 to test with)