Your favorite file system?

ext3
51% (4154 votes)
reiserfs
31% (2490 votes)
xfs
9% (747 votes)
jfs
2% (178 votes)
ext2
3% (222 votes)
other
3% (277 votes)
Total votes: 8068

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.

XFS on file server

Few requirements:

  • quota support
  • exporting via NFS (with quota support(!))
  • simple, possibly on-line resizing (extending)
  • good performance
  • journalling

Let me know if JFS or reiser finally support all these things (when I tried them, there was no quota support and/or I couldn't export it with NFS). Ext3 could be used for my use case but it lacks on-line resizing, what in multi-user environment is not very comfortable. So, obviously XFS was the only and best choice (although it had some nasty bugs - fortunately solved long time ago...). Maybe ZFS will beat it, but others are no match for XFS in my opinion. Big thanks to SGI (I started using XFS since the first version they released for linux, and never regret it...).

Go JFS

JFS is way faster than ext3. I even tested jfs to see if it corrupts when power loss and nope no power loss no fsck. EXT3 needed to run fsck 3 of the 10 times.

Ext3 sucks because of its lack of dynamic inodes and constant fscks.

JFS has extents,dynamic inode allocation and no fsck needed every certain amount of mounts =)

XFS missing the ability to shrink.

I really would like to use XFS but it is missing one essential feature for me which is shrinking. Since I use LVM (mostly in combination with software raid6) on a lot of systems, I sometimes run into the situation where I need to shrink a partition to make more space for some other partition. With XFS this is not very practical so I am forced to using reiserfs or ext3 for most of my systems but I know for performance and other reasons XFS would be a better choice.

REISER4 - THE BEST

REISER4 - THE BEST FILESYSTEM EVER.

.-------------------------.
|.FILESYSTEM.|.TIME.|DISK.|
|.TYPE.......|(secs)|USAGE|
.-------------------------.
|REISER4.lzo.|.1938.|.278.|
|REISER4.gzip|.2295.|.213.|
.-------------------------.
|REISER4.....|.3462.|.692.|
|EXT2........|.4092.|.816.|
|JFS.........|.4225.|.806.|
|EXT4........|.4408.|.816.|
|EXT3........|.4421.|.816.|
|XFS.........|.4625.|.779.|
|REISER3.....|.6178.|.793.|
|FAT32.......|12342.|.988.|
|NTFS-3g.....|10414.|.772.|
.-------------------------.

Column one measures the time taken to complete the bonnie++ benchmarking test (run with the parameters bonnie++ -n128:128k:0). The top two results use Reiser4 with compression. Since bonnie++ writes test files which are almost all zeros, compression speeds things up dramatically. That this is not the case in real world examples can be seen below where compression does not speed things up. However, more importantly, it does not slow things down either.

Column two, Disk Usage: measures the amount of disk used to store 655MB of raw data (which was 3 different copies of the Linux kernel sources).

ZFS

I hope linux add suppert the ZFS (The Solaris 10 `s ZFS is first 128bit FileSystem on the world .Add by Solaris 10 6/06 first.) in furture.

The freebsd core tream had finished it(FreeBSD support ZFS).

Depends on the use case

Which FS I use depends on the expected usage. I mostly stick to ext3 for general purpose use because:
- It's robust and recoverable in the face of many hardware- and software-caused failures. Of course there can be data loss, but the whole FS is not normally totally trashed by relatively small corruptions.
- It's widely supported
- It's reasonably fast except for lots of small files

I tend to use reiserfs, however, when:
- I need high performance with many small files, such as Maildir-like mail spools (in my case for Cyrus IMAPd), mail queues, and so on; and
- The machine's on a UPS and has redundant storage, minimising the chance of FS corruption. Reiserfs is in my experience highly likely to be TOTALLY trashed by relatively small corruption (and I'm talking machine-resets-on-mount-attempt here) and its recovery tools are crap, so it's not a good fs to use on even remotely unreliable storage.

I'm sure XFS has similar areas where it excels, and others where it's quite unattractive. I've never put it to use, but my understanding is that it needs an extremely reliable underlying storage layer like reiserfs does, though for different reasons (zeroing files?). I've never been clear on what its advantages are.

The point is, anyway, that you should consider what you're going to be using that storage for when deciding what file system to use on it.

KillerFS

TOTALLY trashed, but --rebuild-tree effectively does a reload.
Lacks dump so if you want to migrate to another FS then difficult.

I use it on mail server with Maildir store running Exim4 and Dovecot.

I use ctime and mtime for grey list state timers in email filter.
Using a very large number of empty files in a directory.
Per user duplicate message-id detection and various other uses.

I hate databases.

The advantages of XFS is

The advantages of XFS is that it supports huge partitions, it has a real-time volume for streaming media and XFS is well suited for large files. It also supports online defragmentation. But if you are concerned about data integrity then you shouldn't use it as XFS tends to corrupt very easy

where is reiser4

reiser4 has been stable for a long time and has a very large userbase. it's a shame you didn't include it on the list. i voted other

Reiser 4 is *NOT* stable

As a matter of fact, a former employee of Namesys (The company founded by Hans Reiser) called out Hans on the Linux Kernel Mailinglist for "doctoring" the reiser4 benchmarks. Don't believe me? Read the publicly archived thread and the very lame replies from Hans:
http://lkml.org/lkml/2004/8/27/70

Then take a look at some REAL benchmarks showing how reiser4 is a regression:
http://linuxgazette.net/122/piszcz.html

Sorry, I'm not impressed

Ext2 For encryption

I use ext2 on all mobile devices and secure servers as it allows me to run software such as loopaes (blowfish in my case) that way if some llama jacks my shit its useless ;-)

ext3

While I personally use ext3, and I have no problems with it. Some times it can be a little jerk.. lol, so I put other just for the hell of it. :)

Where's the swap file system?

I use that on ALL my servers!

SWAP isn't really a

SWAP isn't really a filesystem. It behaves like one though

SwAp

explanation: slightly autistic people do not readily recognize humor. forgive and forget

What?

It's humour you troll!

ext3 boots from its first sector

While XFS and Reiser can't. XFS lacks good diagnostics/reparation tools.

xfs lacks?

What on earth do you mean by 'XFS lacks good diagnostics/reparation tools.', specifically?

he probably doesn't know

he probably doesn't know about the existence of xfs_check and xfs_repair :-)

I don't know quite what you

I don't know quite what you mean by "boots from its first sector." I use ext2 on the boot partition because journal files are huge (32 MB or more) and I don't use the boot partition for anything except holding kernels.

XFS has cannot fail, so I don't care whether it has diagnostic or repair tools.

Reiser, on the other hand, I have lost data to many a time.

XFS is not perfect you know.

XFS is not perfect you know. The biggest problem with XFS is that its zeroing files with binary zero's if it can't recover them. It's not wise to use XFS if you are concerned about data-integrity. I have lost files due to this zeroing bullshit of XFS

ReiserFS - not anymore

I used to love ReiserFS for the speed. In fact, I would still use it on a system in a datacenter, but not my desktop.

I was using Suse 9.something and lost power to the box. Lost my whole home dir and vmware machines. According to the research I did on the net, the way/when reiser, and xfs if I recall, write to disk and write to journal make them fast, but very intollerant to powerloss.

I went back to ext3 for most things.

RE: ReiserFS - not anymore

Well i never had those problems here with ReiserFS V3. One year ago i did some experiments with ReiserFS just to see how well it protects user data. Main partition on my linux box was ReiserFS V3 mounted with the data=ordered option. I switches off my box 20 times without properly halting it.... you know just hit the reset button :)... I NEVER had any data corruption during this test so i guess somethings wrong with your system, maybe?

reiserfs powerfail test...

I've also done brutal powerfail tests with reiserfs. I'm talking: poweron, run, let machine run, yank plug, plug back in, let it boot half way, yank plug, let it come back, yank, shake, stir, repeat... and repeat... and repeat...

Never lost a thing. The machine just kept coming back up and returning to its journals to bring things back clean.

As for recovering and administration of reiserfs file systems, I've found the tools "reiserfsck" and "reiserfstune" to be handy. :)

reiserfs powerfail test

> I've also done brutal powerfail tests with reiserfs.
> Never lost a thing.
I automated this test.
60 sec ON, 60 sec OFF => even after 24 hours no problem.
But then test-2:
6 sec ON, 6 sec OFF, already within half an hour the system was corrupted.
I repeated both tests with same result.
My conclusion: when reiserfs is initializing it is sensitive for powerfailure.
Any comments or suggestions for improvement ?

reiserfs reliability

I've found reiserfs fairly reliable in general use, and its journalling functionality to work as advertised. A BIG exception is when the underlying hardware is having problems - bad sectors on a disk, occasional data corruption due to bad RAM, overheating CPU, etc. A small corruption to the underlying storage tends to trash reiserfs file systems totally and beyond recovery in my experience.

If you're running good, stable hardware with redundant storage, reiserfs can be quite attractive for some roles, such as maildir-like mail spools and proxy cache directories, where it's small file performance is a big advantage. Of course, you always keep good backups of anything you care about, so even if the worst does happen you're not set back too badly ... and reiserfs should be fine on reliable hardware. Just make sure you test your disks frequently.

In particular, I've found that bad sectors due to a failing disk seem to make reiserfs filesystems essentially unrecoverable a lot of the time, whereas it's normally possible to copy most files off an ext3 fs on failing media. The right answer is, of course, "don't use failing media" but that's rather easier said than done on consumer systems. If you make the mistake of running reiserfsck on a filesystem that's resident on a disk with bad sectors your filesystem is amost certainly toast in my (unfortunately not too limited) experience. Even if you dd_rescue the FS to another disk on another machine then attempt to mount it read-only and copy data off, you may see issues like unexpected kernel panics and even sudden machine resets due to reiserfs's limited error checking. If you reiserfsck the file system to attempt to repair it, don't be too surprised if it's unable to repair the file system or leaves it in a worse state than it started out in.

I had a development server that was panicing regularly and sometimes just rebooting with no warning. I eventually discovered, and attempted to repair using reiserfsck, file system corruption on the machine. This appeared to work, but the issues returned shortly (and worse). At that point I finally realised the disk was failing - it's now the first thing I check using smartctl, but I didn't know disks were evil garbage then - and tried to copy some files off the machine to save some time rebuilding it. I also disabled swap in case bad sectors in the swap were causing the reboots, but this had no effect on them. Copying certain files off the reiserfs volume would just reset the machine. reiserfsck didn't help, nor did using dd_rescue to copy the FS image to another machine. I wasn't at all impressed, espcially since an ext3 FS on the same machine seemed fairly intact except for a few files that'd report I/O errors when I tried to copy them. It was hardly a disaster - this was a development machine after all - but inconvenient and not reassuring about the file system.

Same problem.

mine was a slow, creeping corruption of the filesystem, which I suspect was initiated by one or more power failures (our rural grid is unreliable). Every once in a while, when I make a request to load a file, the hard drive goes into an endless cycle...when you get out of X, you see the dreaded "DriveReady SeekComplete Error (or something like that). As time goes on, it does it with more and more files. The only way to end the loop that I can find is to reset the machin, which I'm sure doesn't help matters. I've had this happen now on two different computers, and I'll only use ext3 now. I've since reformatted that disk with ext3 and it still works marginally, but eventually it will have to be replaced (and run only with my UPS, which was not available to me at the time this occurred).

open a large file of yours,

open a large file of yours, preferably something valuable to you. do something like cat /dev/random >> myfile then pull the plug.

resifer will toast it.

re: open a large file of yours,

So will any file system if you overwrite it with random shit!

not if you are on ZFS

not if you are on ZFS

That's appending, you know.

That's appending, you know.

no it isn't

no it isn't

yes it is

yes it is

and so will JFS and XFS

and so will JFS and XFS

Heeeeey....

I voted Ext3, cause that's what I end up using 99% of the time...

But if there was a lil' more non-linux support for ReiserFS (Specifically, Win2k support: I like being able to snarf stuff from my linux partitions while booted into my gaming machine) I'd be 100% Reiser. Speed increase... ::slobbers all over the keyboard::

visualrfstool

Visual RFStool is your solution. It is rtftool (a usermode program that can read ReiserFS partitions), and a graphical front-end to rfstool (but it barfs on files with spaces in their names). Now, go use your ReiserFS from Windows!

Speed increase??? XFS is

Speed increase??? XFS is faster than ReiserFS in most cases except in very small files

Smaller files are 90% of your system and mine

Every time you use anything on your system, dozens or hundreds of small files are hit. Optimizing for small files (as long as large file access doesn't suffer too much) is the only sane thing to do for most environments, including desktop and webserver usage.

That said, I use ext3 on most of my systems, since I don't want to hassle of going with something non-standard (I use Fedora Core 5 on the desktop and RHEL on my company server). But when I was building web caches with Squid, I always used ReiserFS (and did tons of benchmarks and real world testing seven years ago, and periodically over the years, to be sure I was right and remained right).

Also note that as filesystems progress, small file access becomes more important. Quotas, ACLs, meta-data, etc. all are generally tiny files (or tiny file equivalents, since they're hidden from view).

Optimizing for large files only makes sense on your video or audio storage drives, if you are a video producer or audio recordist. Perhaps also if you do a lot of work with ISOs. I think it's pretty clear ReiserFS is on the right track, though Hans has some issues working within the Linux development process and it leads to a lot less adoption of his team's work than it deserves.

I don't think that Hans has

I don't think that Hans has issues with the Linux developers, it's rather the other way where the VFS people refuse to modify the VFS so that Reiser4 would be included into the mainstream kernel. They are blind to see that Reiser4 introduces something new in filesystems which is plugin-based functionality

read reiserfs from windows

YAReG - Yet Another R(eiser)FStool GUI http://yareg.akucom.de/

Check this little app out to snag files from your reiserfs partition on windows. I only boot into windos to run DartPro, my dart practice program and to update some data in a spreadsheet. I use this to copy the spreadsheet from my /home directory when I've been using it in Linux.

Wish it was a tad more transparent, but hey...it's FREE and it works :)

Reiser FS 4, if it ever

Reiser FS 4, if it ever becomes stable.

It is stable only not

It is stable only not included in the mainstream kernel bcz it breaks the VFS. You can find Reiser4 in the 'mm' kernel though

JFS is the best of those

JFS is the best of those listed.

not if you care about data

not if you care about data integrity :)

what about power loss

Every time my machine with JFS loses power I have to manually do a checkfs and remount it.
Make an rc.local job do check on that? Sure but every daemon that depended on the filesystem has to be checked too. Ugh.

Agreed!

Another vote for JFS.

Bullshit

I suggest you do some research before spewing crap out of your mouth. JFS is far from the best.

Asshat

it doesn't say best. it says favorite. stfu and gbtw.

ZFS!

I refuse to use any filesystem until ZFS is ported to linux (in kernelspace)! 128 bit storage pooling with snapshots... *drool*. Don't ask me how I'm posting this comment from a computer with no filesystem, it is a very long and complicated process that I do not want to get into at the moment.

ZFS is the future. If only

ZFS is the future. If only more OSes supported it well.

union

unionfs - totally underappreciated and underutilized

it's not a FS in its own

it's not a FS in its own right, but a layer to make an union out of other filesystems. I tried to use it once, but in my case it wasn't a good choice because every time a file was written to, it copied the whole file over instead of only the parts that changed. I ended up using cowloop, which works at the sector level.

Vote for NTFS!

And embrace and extend :-)
NTFS files run better on Linux.

Oh yea? did you benchmark

Oh yea? did you benchmark NTFS agains XFS as I did?

Where is NTFS?

I use NTFS daily, where is it?

Stupid poll

ya?

do you defragment your drive daily too?

No...

*You* are the stupid one. Go away, M$ troll.

Linux site!

If it's not obvious, this site is about Linux OS.

I mean, like you have a choice on file system on Windows, it's NTFS or prehistoric FAT. ;)

i like ext3 because.....

its journalised and can now be read by xp, see
http://www.fs-driver.org/

The next version of reiserfs, not the current one.

The future vision of reiserfs is the attractive part of this one. It's one of the bigger innovative computer advances on the near term horizon.

not when compared to ZFS :)

not when compared to ZFS :)

zfs rules

Try Sun's zfs and you will see that those fs and your vintage fdisk are really out of date.

Not runnin' Solaris

Well, since the only stable implementation is in Solaris, it's not something I can "try out", I know that it's getting ported to FreeBSD, and I'm looking forward to that, but I innerly hope that Sun will dual license ZFS under the CDDL and GPL so that it could be a very good option in Linux.

BSD

it'l rather be a bsd style license (if any oss license) if it gets incorporated into freebsd.

Cheers,
-S

not sure

does using a reasonably new Linux distro garenty that you are not using ext2 ?

I know I am using ext2 or ext3 but I do not know which.
Gentoo 2005.1 - 2.16.17 on AMD64

check your fstab entries :)

check your fstab entries :)

Incomplete list

ZFS should be mentionned !

NO ZFS SHOULD NOT BE MENTIONED.

The url www.linuxinsight.com kinda makes me think they are talking about Linux filesystems. (Just a guess).

ZFS will probably never make it to a real distribution that actually adheres to the opensource licencing, because the CDDL is not compatible with the GPL.

Sure it can be written, but I dont think anyone in any sane distribution is going to risk the sun legal team coming down on them hard during this time of hardship.

how about that? they are

how about that? they are porting ZFS to Linux on the FUSE project and it's 100% GPL

Digg it!

Digg it, so that more people know about the poll. More votes - more correct result!

more vote doesn't mean

more vote doesn't mean correct result...

Hm...

No poll in the world is completely correct or thrustworthy, but more diverse public and more votes definitely take it closer to the truth, IMHO.

There's another poll on the site where the topic is databases. That one received much less traffic and by looking at the results you can notice something strange. PostgreSQL is more popular than MySQL? Of course not, it's just that results are skewed with few popular articles on this site. The results would be very different (and more correct!) if that poll received similar amount of traffic like this one.

Comparison of file systems

Wikipedia has very nice comparison of file systems.

no GFS?

I chose ext3 b/c it's available on all my machines and I use that the most. But a few of my boxes have GFS... now that's some good stuff.

No mention of tmpfs

I do most of my work on tmpfs, and commit the results to an svn repository on ext3. I have one tmpfs on /dev/shm and bind subdirectories of that at /tmp, /var/tmp, /var/run and /var/lock. There are several reasons for this.

I said some very rude words when I found a PID file had survived reboot and was causing confusion. I said similar things when my serial port was 'locked' for a similar reason. This has not happened since I used tmpfs.

The files most at risk from power failure are the ones you are working on. What is more, they pose the greatest threat to the rest of the filesystem. I have never had half-written files on tmpfs damage my home directory ;-).

I used to use reiserfs-3, but I wanted xattr's. I am curious about reiserfs-4, but I will not use it for a long time. Last I heard, the VM changes provided functionality only for reiserfs. If the changes are worthwhile, they should be available to all filesystems. While changes are filesystem-specific, there is a bad risk that people not using that file-system could unknowingly break the file-system specific parts of the VM. That is something I really do not want to suffer from.

If filesystem performance actually made any difference to me, I would try jfs/xfs and also try a high speed flash disc. I would expect that the fastest file-system depends on the benchmark, so the only benchmark that matters is your own usage.