Which FS do you prefer?

[Flatline]

This is primarily a 'nix question, but any old file system is, of course, up for discussion.

I have been leaning toward ReiserFS recently, but I was wondering if anyone had a specific preference; any explanation of why the preference exists is more than welcome.

I was just hoping to have a thread on the merits of various file systems (primarily the journalized file systems), so please NO FLAME WARS!!

 

[Goosemaster]

XFS was SUPPOSED to be good.....now it seems that corruption and missing files are reglar problems...

ext3 baby.......I've never gotten it to work, but heard great things

 

[pac1085]

I've been using XFS for about 6 months with no problems whatsoever. So far, I'm happy with it. If something comes up, I'm gonna try ReiserFS. Alot of people say its great. I run EXT3 on my Debian web-server. Haven't had any problems with that either.

 

[Flatline]

I've had a couple of gurus tell me that ReiserFS is quicker than EXT3 with small files and EXT3 is better with larger files, but it's difficult to find any hard evidence...

Anyone tried JFS?

 

[Darkcirc]

I have seen some benches on ext_2_, reiser, and XFS. It would seem that ext is faster than Fat32, reiser is extremely to a little faster than ext2 and XFS is faster than reiser. I have heard the corruption complaint about XFS a couple times but it all seems to pertain to power failure situations. I have had such power issues but everything seems to come back up just fine. I've used ReiserFS extensively, and like it, I havE almost complete confidence in its ability to come back up. I have toyed with ext3 and I'm not impressed, seems like the MS FAT type upgrades, slap another layer on top and call it new. You can mount ext3 as ext2 and read and write just fine, it is simply a journal kept in God knows where that makes it '3'. Summary: XFS for the workstation, ReriserFS for the server, ext2, FatX for things you don't use EVER. And yes currently the benches I speak of are directly from my ass, I can't remember where I got em from, but it was research on this exact same problem for Production servers @ school, before XFS was supported by linux fully and *right* after reiser got Module supprt. Hope this helps.
dc

 

[Sunner]

XFS user here.
I've only recenely seen some reports about problems with it, but nothing hard though.
ReiserFS OTOH has had alot of problems, and while it seems like they're ironed out now I'll stay clear of it for a while longer while it matures.

Ext3 does have some performance enhancements compared to Ext2, which I suppose could at least make it Ext2.5

 

[thornc]

I use ext3... but then again I haven't worried about any performance issues.... I just use it because it is backwards
compatible to ext2!

 

[n0cmonkey]

FFS + softupdates

 

[Derango]

I use reiserFS everywhere. (/boot is still ext3 though) No problems.

 

[Electrode]

XFS

I've had problems with corruption and file loss on ReiserFS, stability problems with JFS, and Ext3 is an ugly hack that I just don't trust.

 

[TheOmegaCode]

ffs, ifs, devfs, nfs, cifs, and ext3, oh my!
I'm looking into using XFS on my next box.

 

[Toxic]

I'm using ReiserFS. No problems.

 

[Nothinman]

XFS all the way. I use ext3 on my non-x86 boxes because XFS hasn't had much Linux Alpha or Sparc bang time and I don't want to be the tester just yet =)

ReiserFS OTOH has had alot of problems, and while it seems like they're ironed out now I'll stay clear of it for a while longer while it matures.

Agreed. I also don't like their 'development model', each revision of reiserfs has had on-disk format changes so far and reiser4 is supposed to be a complete rewrite, doesn't sound too promising if you ask me. It sounds really good and if they ever get figured out what they want to do, it might be the next defacto-standard Linux filesystem, but at this rate it'll be a while.

 

[Darkcirc]

Correct me if I'm wrong but reiser seems to be goig the way of the OS/390, a gigantic relational database type Layout. Don't get me wrong, I like reiser, I still use it in a lot of boxen, I just think the 'new direction' is idealistic rubbish, seems to want to 'shake the foundations of linux' but I doubt it will be greeted with open arms. I for one will be the guy standing there arms crossed waiting for them to make it with the enterprise crowd. And yes reiser can be a little freaky when upgrading, and it seems to leave alot of stuff that other FSes will ignore, for instance, mkreiser, then mkfs.xfs and mount the part, allow linux to choose, in my experience it still chooses reiser and fscks like a mofo, baffles my mind and destroys data like a champ, and I don't know who to blame. Either way, IMHO xfs is a great FS, I'm trying to move everything to XFS, but I must admit this loosing files has me a little on edge... ::runs off to reformat and restore from tape::. Ahh, good thing there don't seem to be any hardcore softgeeks here, spouting about addressing arrays, and node arrangement or this'd be a hell of a religious war. Seems to me that the tide goes reiser/XFS, so at least you have it narrowed down. and I still think ext3 is a patchjob and using it encourages recycling of old stuff, winders9x anyone? 32bit code with 16 bit addressing? FAT was only ever ment for floppies, hence Fat12, Fat16under512 and FAT16 over 512(i think its 512), FAT32, and FAT32.2...
dc

 

[BlackOmen]

Using ext3 right now. Electrode was right when he mentioned above that it is an ugly hack, but for some reason I trust it.

I chose ext3 because of the following reasons:
XFS + the preemptible kernel are mutually exclusive right now. Once some of the preempt issues get ironed (possibly already ironed out in 2.5), I'll probably use it.
JFS: it's currently the butt child over on the Gentoo forums, but their preferences change daily. Once I hear more real world results with JFS, it might be an option.
ReiserFS: the data corruption issues from early 2.4 kernels are still leaving a bad taste in my mouth. Also, when the default options don't work, I begin to get a little concerned. I'm referring to the recommendation to specify notail as a mount option. This may have changed, but I've long since given up on Reiser.

 

[n0cmonkey]

Originally posted by: Darkcirc
Correct me if I'm wrong but reiser seems to be goig the way of the OS/390, a gigantic relational database type Layout. Don't get me wrong, I like reiser, I still use it in a lot of boxen, I just think the 'new direction' is idealistic rubbish, seems to want to 'shake the foundations of linux' but I doubt it will be greeted with open arms. I for one will be the guy standing there arms crossed waiting for them to make it with the enterprise crowd. And yes reiser can be a little freaky when upgrading, and it seems to leave alot of stuff that other FSes will ignore, for instance, mkreiser, then mkfs.xfs and mount the part, allow linux to choose, in my experience it still chooses reiser and fscks like a mofo, baffles my mind and destroys data like a champ, and I don't know who to blame. Either way, IMHO xfs is a great FS, I'm trying to move everything to XFS, but I must admit this loosing files has me a little on edge... ::runs off to reformat and restore from tape::. Ahh, good thing there don't seem to be any hardcore softgeeks here, spouting about addressing arrays, and node arrangement or this'd be a hell of a religious war. Seems to me that the tide goes reiser/XFS, so at least you have it narrowed down. and I still think ext3 is a patchjob and using it encourages recycling of old stuff, winders9x anyone? 32bit code with 16 bit addressing? FAT was only ever ment for floppies, hence Fat12, Fat16under512 and FAT16 over 512(i think its 512), FAT32, and FAT32.2...
dc

In reference to the part I bolded, is that directed at me?

I think softupdates sounds like a better technology and in some papers it does quite well. I spoke my preference near the beginning of the thread and see no reason to start a flamewar Q Yes, I said that ). I am impressed by some of the journaling filesystems, and the last time I "tortured" myself with linux, I used XFS. Unfortunately the machine wass overcome with hardware problems, filesystem corruption, and who knows what else. When I get the machine back up, XFS will yet again be loaded. If you want a flamewar, I can try to give you one. But this is definitely not an area of expertise for me, and although it could be fun, it is not necessary

BSD rules.

 

[n0cmonkey]

Originally posted by: BlackOmen
Using ext3 right now. Electrode was right when he mentioned above that it is an ugly hack, but for some reason I trust it.

I havent heard too many bad things about it yet. But I dont typically keep up on linux tech.

I chose ext3 because of the following reasons:
XFS + the preemptible kernel are mutually exclusive right now. Once some of the preempt issues get ironed (possibly already ironed out in 2.5), I'll probably use it.

I am under the impression that the preempt patches help out desktops more than the typical server. Whether I am using a system as a desktop or not, I prefer to keep it as similar as I can to my servers (+X ). So preempt was never a big deal to me.

JFS: it's currently the butt child over on the Gentoo forums, but their preferences change daily. Once I hear more real world results with JFS, it might be an option.

IBM is behind Linux pretty well. Last I had looked at JFS it was not ready for primetime, but I doubt it will stay that way for long. One of the Linux mags I picked up recently had an article about the different journaling FSes, and it seemed to speak highly of JFS.

ReiserFS: the data corruption issues from early 2.4 kernels are still leaving a bad taste in my mouth. Also, when the default options don't work, I begin to get a little concerned. I'm referring to the recommendation to specify notail as a mount option. This may have changed, but I've long since given up on Reiser.

ReiserFS just changes too much, too quickly for my tastes. Evolution over revolution... But Im a BSDer, Im used to slow.

Not dissing anyone's choices, just my opinions.

 

[Abzstrak]

JFS was very nice before the port to linux, I haven't used it yet in linux... I'm currently running reiser on most things, ext3 on the rest. I've never had a problem with reiser, but did have a few issues with XFS. Reiser also feels snappy when opening things like gimp with a bunch of plugins -- just opens up noticably faster.

 

[Flatline]

Yeah, I have noticed a speed difference between EXT3 and ReiserFS on the same machine (test box at work) as well; Reiser seems to be quicker with most things than EXT. Anyone have an idea which is faster when recovering?

Some 'nixes (RedHat, noticably) don't really give the option of formatting a partition with Reiser and a couple of other file systems (without some nasty hacking); anyone know why?

Noc, do you remember which 'nix mag you were looking at that had that article? That would be interesting (boy, I just showed how geeky I am, huh)

 

[BlackOmen]

Using ext3 right now. Electrode was right when he mentioned above that it is an ugly hack, but for some reason I trust it.

I havent heard too many bad things about it yet. But I dont typically keep up on linux tech.

I've been reading some usenet (linux.kernel a lot), and there have been a few patches to fix ext3 corruption recently. I havn't experienced any myself, but I guess it's that typical internet horror story that because it happens to 1 person, it will happen to everyone.

I chose ext3 because of the following reasons:
XFS + the preemptible kernel are mutually exclusive right now. Once some of the preempt issues get ironed (possibly already ironed out in 2.5), I'll probably use it.

I am under the impression that the preempt patches help out desktops more than the typical server. Whether I am using a system as a desktop or not, I prefer to keep it as similar as I can to my servers (+X ). So preempt was never a big deal to me.

Preempt does benefit desktops more than servers. As a matter of fact, while increasing responsiveness, it may decrease throughput. Definitely wouldn't want that on a server. When using the preempt patches, I notice that rendering times in Gimp have gotten longer. But my machine does grind to a halt either.

The problem with xfs + preempt was that it would cause processes to exit with a preempt count > 0. If the count is greater than 0, something is wrong. I havn't been able to find a clear cut answer as to what causes the problem specifically. I never used that combination myself, however lots of people over on the Gentoo forums had a problem with it.

 

[Sunner]

Originally posted by: Flatline
Yeah, I have noticed a speed difference between EXT3 and ReiserFS on the same machine (test box at work) as well; Reiser seems to be quicker with most things than EXT. Anyone have an idea which is faster when recovering?

Some 'nixes (RedHat, noticably) don't really give the option of formatting a partition with Reiser and a couple of other file systems (without some nasty hacking); anyone know why?

Noc, do you remember which 'nix mag you were looking at that had that article? That would be interesting (boy, I just showed how geeky I am, huh)

Not that I've asked them, but I get the feeling the reasons are the same as have been diwscussed in this thread.
That is, it's still too immature, and they're waiting for it to become more tried and true.
Say what you want about Ext3 just being Ext2 with some features tacked on, but it's definately a very stable FS, and considdering RedHat's focus on corporate users, that's a very important thing to considder.

 

[Flatline]

I always kind of assumed that their architecture had perhaps been optimized for a particular FS, but I've never seen anything to discount that or support it (just a hunch, I guess). It would make sense that RedHat would default to EXT3 because of their long history with the EXT file systems, but not making some of the others available at all seems like overkill...

 

[Nothinman]

Correct me if I'm wrong but reiser seems to be goig the way of the OS/390, a gigantic relational database type Layout.

Is that his plan this week? =)

I was just thinking his idea of filesystem plugins was new and should allow for a lot of nice features.

I'm trying to move everything to XFS, but I must admit this loosing files has me a little on edge.

In all my time of using XFS (since the 1.0 release) the only files I've lost were being written or still had unflushed dirty buffers during a crash, I've never had anything 'randomly' lost.

and I still think ext3 is a patchjob and using it encourages recycling of old stuff, winders9x anyone?

It is a patchjob, mainly to allow users an easy transition to a journaled filesystem. A lot of people understand ext2 and would like to keep it, but also need to avoid running fsck on hundreds of gigs of filesystems, ext3 gives them that.

I am under the impression that the preempt patches help out desktops more than the typical server

Depends on the server load. The preempt patch can help data throughput in many cases because it allows the kernel to schedule other things while waiting on I/O where it would normally block everything.

JFS was very nice before the port to linux, I haven't used it yet in linux...

You realize it's the JFS from OS/2, right? They havn't incorporated all the AIX features yet.

while increasing responsiveness, it may decrease throughput

Not according to RML. Of course it's dependent on a lot of variables.

 

[CTho9305]

since you allowed non-*nix filesystems, I have to say i've had only good experiences with NTFS. No data loss (except when I encrypted stuff and lost the key ), good performance, and ACLs (for good permissions handling)

 

[Flatline]

NTFS is OK for desktops (for large server systems, I want a journalized file system). The early versions were a bit of a nightmare, but it's hard to tell whether it was the file system of the OS that was the problem (NT 3.x was horrifying)