ReiserFS vs XFS vs JFS vs Ext3 vs Ext2

[marat]

I read a lot of stuff about different Journal filesystems, but I'd like to hear some examples from real people. I have ReiserFS working on my Linux box and it fine so far (little bit slower than ext2 though). What is your journal(or not?) filesystem of choice in terms of:

1. Speed
2. Compatibility
3. Stability
4. Features
5. Space usage
6. Advantages vs Ext2

I am going to run it on Linux box with > 2.4.10 kernel. Suggestions?

 

[Nothinman]

XFS, I believe it to be the most mature as it's been being used on Irix for years and it's userspace tools are light years above anything reiserfs has. It also supports things like ACLs, quotas, external log devices and generic Extended Attributes. The only thing reiserfs does a little bit faster is deletes.

JFS is a port of the OS/2 code so it's still missing a lot of the "enterprise" features of JFS on AIX. I havn't used it so I don't know how stable it is.

I also use ext3 on one of my boxes because it was an easy upgrade from ext2, it supports full data journaling and XFS isn't as tested in 64-bit sparc.

 

[marat]

2Nothinman: Thanks, I'll try it.

Anyone else has goof/bad experience with any FS?

 

[SPB]

ReiserFS is REALLY GOOD. Never had a single problem with it using Mandrake 8.1. The default file system (I think it was ext2) was more problematic. As for the technical side of things I couldn't tell ya and I didn't use any of the other file systems.

 

[n0cmonkey]

Ive been wondering about this myself. I prefer softupdates, but Im a BSD bigot. ext3 and XFS look like the best. But I really dont know more than hearsay. I would choose XFS because of the fact it has some real world testing.

 

[mcveigh]

<< Ive been wondering about this myself. I prefer softupdates, but Im a BSD bigot. ext3 and XFS look like the best. But I really dont know more than hearsay. I would choose XFS because of the fact it has some real world testing. >>



I agree, I haven't used XFS yet but like what I've read, EXT3 is nice b/c you can upgrade a ext2 partition to it.

 

[Armitage]

I had significant stability problems with ReiserFS under Mandrake 8.1 They cleared up as soon as I rebuilt the system on ext2.
That was an early 2.4 series kernel, and many ReiserFS problems were found subsequently, so for 2.4.10 you may have better luck, but overall, I think ReiserFS is still a bit green.

Right now I'm running ext3 and its been working great, but I haven't had the misfortune to have to exercise the journalling capability yet.

 

[marat]

<< I had significant stability problems with ReiserFS under Mandrake 8.1 They cleared up as soon as I rebuilt the system on ext2.
That was an early 2.4 series kernel, and many ReiserFS problems were found subsequently, so for 2.4.10 you may have better luck, but overall, I think ReiserFS is still a bit green.

Right now I'm running ext3 and its been working great, but I haven't had the misfortune to have to exercise the journalling capability yet. >>



So what happened? Madrake 8.1 uses 2.4.8 kernel. There were some issues that were cleared for ReiserFS in 2.4.9, but... Can you elaborate?

 

[BlackOmen]

Since the 2.4.15/2.4.16 kernels have ext3 support as a standard kernel option, I have converted all but my / partition to ext3. All partitions survived a deliberate power off, after about 10 days of uptime and heavy usage.

XFS is supposed to be one of the faster in that it uses a better b-tree indexing. For some really nitty gritty info, check out this. AFAIK any kernel version can be patched to XFS.

As far as compatibility, I chose ext3 because if things didn't work out, a cleanly unmounted ext3 partition can be mounted as ext2. Ext3 is also the easiest to upgrade to if you have the latest e2fsprogs and the latest linux-utils (I can find exact links if you like).

JFS is still pretty young, and only specific kernels can be patched to it. Don't know much more than that.

Reiserfs was buggy in earlier 2.4.xx versions, but I know people that swear by it. It is also the fastest journalfs in write times, where xfs is the faster with read times. There is also native kernel support for reiserfs.

 

[fivepesos]

somehow a few power outages killed my router running XFS. it would give me kernel panics. weird, i never did get it fixed. i got pissed and went to back to freesco (router on a floppy for you guys who dont know), nothing like a writeprotected floppy to keep your filesystem safe

i currently run ext3 and im very happy with it. no fsck. thats about it. performance is the fairly similar to ext2 but ive only stressed the FS by doing typical desktop stuff (ie move my cd image archive between partitions, while compiling a kernel, while playing vcds). so i guess if thats what you want, ext3 is good enough.

XFS is supposed to be faster than XFS, but ive hardly noticed, but i havent been looking carefully. ive found reiser to be very mature and it comes prepackaged with many distros installation, suse and redhat for example.

 

[Armitage]

<< So what happened? Madrake 8.1 uses 2.4.8 kernel. There were some issues that were cleared for ReiserFS in 2.4.9, but... Can you elaborate? >>



I haven't tried ReiserFS again since Mandrake 8.1 Maybe the problems have been solved, but I'm happy with ext3 for the moment.

I posted to the Mandrake support mailing list. Several others described similar problems, and the reply from the mandrake people was "bad hardware". Which could be I suppose. But It's run ext2/3 under heavy cpu & io load now for several months without any significant hiccups (other then some KDE stuff ) Subsequent to this I saw a fair bit of discussion on LKML about problems with ReiserFS that were cleaned up around 2.4.9/10 I believe.

My setup is a little nonstandard. RAID 1 on a 3ware card. That might play into it.

 

[marat]

Thanks for your info, guys.

 

[DaHitman]

I have been using ext3 in a production environment since redhat 7.1 came out and it has run flawlessley.... I even converted the ext2 to ext3 on the fly while it was all up and running...very cool.

 

[Nothinman]

ext3 is cool, and it's full data journaling (not on by default) gives added protection to the data on the disk and not just the filesystem structures which is nice but slow.

 

[bubba]

Can you elaborate on that Nothingman?

 

[Nothinman]

A journaling filesystem keeps a log of everything it does to the filesystem, that's how it avoids fscks when it's unmounted uncleanly, but most only log changes to filesystem metadata (inode information, superblock changes, etc) because it's fast and it still keeps the filesystem in a workable state in the event of a crash. But this means that in a crash even though thet filesystem itself is kept consistent because of the journaling the files themselves could be lost. ext3 has an option that you can have it log everything even the file data, which is slower but you have a much better chance of having your files be the way you expected them when then the unclean umount happened.

 

[bubba]

Thanks Nothinman. So, how does one enable full data journaling? How much does it affect performance?

 

[Nothinman]

So, how does one enable full data journaling? How much does it affect performance?

With the mount options journal=data I believe, I don't have it in front of me to look though.

I don't use it on my main machine so I can't say how noticable the performance drop would be. The box I have it running on only does web and mail and the mail is only for me, so performance isn't that important.

 

[StuckMojo]

according to the ext3 whitepaper by redhat, the speed difference is minimal, so you might as well
go for the data journal setting, unless speed is of utmost importance to you, like in a database
server or something.