“The conclusion is obvious by the “Total Time For All Benchmarks Test.” The best journaling file system to choose based upon these results would be: JFS, ReiserFS or XFS depending on your needs and what types of files you are dealing with. I was quite surprised how slow ext3 was overall, as many distributions use this file system as their default file system. Overall, one should choose the best file system based upon the properties of the files they are dealing with for the best performance possible!” Read the whole article at the LinuxGazette.
Benchmarking Filesystems: JFS, XFS, ReiserFS, ext3
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
And in the end, it’s still the same result. Pick whichever one you want, cause there is no clear best.
I would like to see benchmarks of Reiser4 though. It is apparently very close to release.
What i really would like to see is the same thing compared with freebsd ufs2, microsoft ntfs, macos hfs(?) and so on
That would be unfair as you can’t run all those in Linux. Switching out kernels introduces a second variable.
There are too many variables involved when comparing file systems in different operating systems, paticularly when you make a jump between hardware platforms (ie. you could not use the same hardware).
Unless you want the safety features of journaling, I don’t see a reason to use ext2 at all. Ext3 comes up short in too many areas although I thought that it had been improved to within striking distance of the other journaled FS.
I’ve had minor problems with XFS but what I would have liked to see would be the result of using multiple FS simultaneously.
The one time I tried that – 4 partitions, each with a different FS , on Mandrake 8.1, the performance of the entire system was horrible.
At least this guy used the latest kernel and everything. There’s always someone who clamors about using a kernel that’s 2 bugfix versions old.
But yeah, I think he was trying to make small pictures to reduce bandwidth, but at this point, he’d do best to just can the graphs altogether; they’re completely illegible.
The fscking tarball is only 630 kb; it decompresses to just under a megabyte
He wonders why EXT3 is used by so many distributions… Perhaps it’s because EXT3 has a proven track record and has been in the mainline kernel for some time, unlike other journaling file systems. In addition to the fact that EXT3 filesystems are backwards compatible with EXT2, meaning you can mount EXT3 FS as EXT2. The performance is “good enough” for most people and it’s reliable. That’s enough for me until Reiser 4 becomes stable.
Linux Gazette policy did not allow me to post bigger images.
They specifically made me reduce the images to be smaller, originally they said 160×120.
Here is what I wanted to post (big images):
http://209.81.41.149/~jpiszcz/index.html
Agree with Shawn.
I’ve used Ext2/3 as long as I have been running Linux. Must say, it is absolutely rock solid, never lost any data or had any disk crashes in 8 years I’ve been using linux boxes! (Watch my machine die tonight! 🙂
And based on the tests, apart from the fact it’s not often I create 10,000 directories 🙂 , I’m willing to sacrifice a couple of seconds performance here and there in exchange for the piece of mind that my data integrity is proven!!!
It would be interesting to see it also in comparision (as it should be significantly faster than ext2/ext3 for large directories).
I have returned to EXT3 after several destroyed reiserfs partitions…
reiser is faster, but I agree with Tarball, I’m also willing to sacrifice a couple of seconds to know that my data is safe.
Here is what I wanted to post (big images):
http://209.81.41.149/~jpiszcz/index.html
Thanx for the comparison Justin, wow what a difference reading the article on the link mentioned above
They all seem quite close, but JFS seems to be the overall best performing. Spiffy. Really though, none of them are “bad.” I’d be curious how 2.6 fairs on the same tests…hmmmm.
I’ve lost 2 entire installs using ext2 in the past (power outages). One was Slackware, and the other was Redhat 7.1. I wouldn’t trust it unless you have a UPS. XFS on the other hand, hasn’t bitten me once, even when I was troubleshooting buggy nvidia drivers and constantly locking up my system. YMMV.
Almost 2 years ago at work we started converting our 3ware raid arrays from ext3 over to reiserfs. We started with our 1.2T 8 drive arrays. At first we thought ext3 was buggy but the found out we had intermittent power problems. Anyways, we haven’t had a single problem with reiserfs, I believe we have something like 15 arrays, or ~20TB worth on reiserfs.
One huge thing we love about reiserfs (which is why we kept it on the drives) is that the integrety check is VERY fast, about 40 mins to check 1.2T. With ext3 you can only run a full blown file system fix on it. Now to actually fix a reiserfs partition (rebuild the tree) takes the same as the single check with ext3.
Another thing to keep in mind is how much CPU overhead there is for each filesystem. On a server it probably doesn’t matter much, but on a desktop you will notice the hit. I’ve found reiserfs uses the most cpu, ext2 obviously the least. Jfs seems to have the best of both worlds imo.
I thought ReiserFS4 would be in 2.6.6. judging from this
http://kerneltrap.org/node/view/2761
but maybe 2.6.7.
I look forward to trying it. Because my experience is that ReiserFS3 (on SuSE) is “smoother” and slightly faster than than EXT3 (on Fedora).
Everybody says that EXT3 is more stable. Where does that info come from? More than one original source? EXT2/EXT3 has had the most exposure but that does not necessarily mean that it is more stable or that other FS’s are unstable or not stable for common use.
On the other hand, an unstable file system has got to be the worst experience – especially if the problem does not show up right away.
Its Reiserfs NOT resier
Just that..in the leyenda
which options were used when mounting the filesystems ?
if none were used, it would be instesting to see filesystems benchmarked with common optimized settings (ie: noatime for ext2/ext3, noatime,nodiratime,notail for reiserfs).
this could give a different (and more realworld) picture !
i tend to use reiserfs in my servers, no problem so far, good performance especially with squid cache_dirs where ext2/ext3 performance is just bad. but i read that on AMD64 reiserfs isn’t perfectly stable yes. never tested xfs neither jfs.
It would be nice to know. Ext3 has different modes which vary from the minimum journaling that some other filesystems use to trade speed for reliability to a full featured mode which you can use if reliability is the most important goal.
I’m not saying that ext3 is the best for everyone in all situations but if you want a filesystem with proven reliability on Linux than ext3 is a very solid bet.
Mount options, no special options, mount /dev/hdg1 /mnt.
Perhaps an obvious question. Just to be sure, were there any programs running on the background like imon or famd? What was your kernel config, were there any tweaked settings in sysctl? I know, it might be all default and obvious, but just to take any hesitations away…
I dislike ext3’s slow performance and ridiculously long fsck times but it does have the advantage that you can mount ext3 partitions inside windows. No other *nix filesystem has that capability IIRC.
I’d love to switch over to reiserfs or jfs but I need cross-platform compatibility.
I’ve been using the current version of Reiser4 on my root partition for two weeks now and I must say it works great. I haven’t done any benchmark yet, but with freedesktop’s xserver and Reiser4 everything works faster than ever.
i think it could be interesting to run a second round of benchmark with common performance-oriented mount options (without sacrifying the liability) for each filesytem..
(like disabling acces time updating, notail for reiserfs).
after all, anyone interested in getting the best performance from their hard drives will or at least should use them.
we could see the impact of these settings on each FS. and if it changes the picture.
IMHO somthing bad happened with Hans Reiser & Co near Sept-Oct. 2003, there was some activity burst in lkml and benchmark section on http://namesys.com. But now fs is not ready. If you remember endless “Reiser4 due…” changes and other “next week merges” it is all look strange. I think there was big regression in some internal test that Hans silency keep trying to fix ? Or it all about money ? For some stupid reason I was wait for Fedora Core 1 because for some reason I hope it will based in Reiser4. Damn. Now even main kernel tree does not support it, don’t speak about feature complete FC2 test3.
BTW, recently I search net about Reiser4, ext3, etc. benchmarks… Nothing new since sept/oct 2003, so nobody in universe care about greatest ambitious open project ? I want latest, fastest, strong scientific based fs on my desktop NOW!
Pleeease! Your spend a lot of resourses, work, time, don’t give up!
Reiser may be fast as hell, I don’t trust it anymore.
After loosing (very fast, too) some data partitions, I will never, never give it a try anymore, no matter of the benchs, no matter if it is in the kernel or not, not matter if you pay me for it.
I want a reliable filesystem. I need to know that if there is some problem the downtime will be minimal and the data integrity preserved.
Ext3 may be slower, true, but much more reliable in my eyes.
XFS is another excellent alternative.
I lost my partition to reiser too. So I’m now again on ext3. As I’m not running a fileserver I’m quite happy.
Thanks for the benchmark!
Just like other posters here, I’ve had quite bad experience with ReiserFS, (which is default on SuSE, my first Linux): bad fsck-behavior after pc crash, and even complete loss of data without obvious reason.
So I switched to ext3 and never had any problems since then (about 3 years). In every day use (just a home family pc) I don’t see much difference in speed. And the warm and cozy feeling of safe data is worth a few milliseconds or even seconds of waiting…
Anyways: Experience with JFS, someone?
Cheerz Lars
But one has to keep in mind that ext2/ext3 is probably the most stable filesystem available for linux.
I love JFS, I’ve used it in my desktop box for some time, but youre almost always screwed if your power goes out or your computer crashes unexpectedly. I would recommend jfs in a production server environment however, since server boxes are likely to be using ‘stable’ (often older) packages and would be looking to optimize the system in any way possible. Certainly using JFS would be a good way to do this is you have a solid system.
AIX, I believe, is now using JFS2 and the version of JFS we got in Linux was actually ported from OS/2 I believe. Whats new or different in JFS2 vs what we get in Linux? Anyone know?
I always use either ext3 or reiserfs. depending on which linux distro I am installing, on which type of machine.
let me elaborate…
I like reiserfs for the speed, and I have never lost data with it. however, I have heard the rumours. so for mission critical machines, or friends moving to linux, I will install on ext3.. just in case
I read that comparison so decided to experiment… I made a clean install of mandrake 10 on xfs, on this machine here, and immediatley after the first boot, I hit the power off button. look, it still works. it restarted with no problems at all.
and… it is FAST I am impressed
I will consider xfs a bit more in the future
So far I’ve heard:
EXT2 bites on power outage.
RaiserFS bites on power outage.
JFS bites on power outage.
Now I wonder:
Does EXT3 bite on power outage?
Does XFS bite on power outage?
Or is NTFS the only real secure File System on power outage?
>Perhaps an obvious question. Just to be sure, were there >any programs running on the background like imon or famd? >What was your kernel config, were there any tweaked >settings in sysctl? I know, it might be all default and >obvious, but just to take any hesitations away…
No programs were running in the background like imon or famd or anything else that would affect the benchmarks, crond was also killed for the duration of the tests. There are no kernel tweaks as I wanted to provide a neutral test base, ie, what you or I would get with a stock kernel.
Ext3 journal option was run in the default mode, as to what the default mode is with 2.4.26 & latest ext3 utils I am not sure.
I’ve been using it since before it was including in 2.4 main when a few distros started offering it very early, like Mandrake, and I’ve /never/ lost data to it. Which isn’t to deny that some have, but I can vouch for its reliability. Its never burned me. Ever.
they aren’t rumors. the power went out on one of my machines, and reiserfs totally shat the bed on me. had to reinstall (and used ext3 the next time!).
and fyi i tried everything, reiserfs refused to repair itself using my gentoo livecd.
“That would be unfair as you can’t run all those in Linux. Switching out kernels introduces a second variable.”
Well, I don’t care for that. I don’t want to see HFS, UFS2 and NTFS performance in Linux.
I want to see how good UFS2 performs on freebsd, HST on macosx and NTFS in win xp/2k3. And to get the same hardware freebsd and linux tests could be made on both ppc and x86 so it’s easily comparable with ntfs/x86 and hfs/ppc.
I just want to know how the different oses filesystems performs compared against eachother.
My personal experience so far has been that EXT3 and XFS are the only filesystems that really survive the sudden power-off well.
NTFS does a very respectable job in many cases, but it seems to do synchronous writes rather than journaling. I’ve noticed that the disk check in NTFS can take very long sometimes (NT, XP, Win2K seem to do their CHKDSK in the background), which can bog down the machine. Generally, the filesystem issues are cleaned up pretty well, but if the OS was dealing with a system file at the time of the power-down (the registry, a driver, a library, etc.), you are SOL. On the rare occasion I’ve gotten some truly inexplicable filesystem curruption under XP in that circumstance.
Just to counter some of the posts here, which suspect ReiserFS is not stable enough: I use ReiserFS now for more than 3 years, with linux versions 2.2.x and >= 2.4.1. I used it with a few harddrives and on many partitions, on different Computers. Some of the partitions I resized with the reiser-utils after I began to use them.
I never had problems with it. Not only it survived several power-switch-offs and random power outages, but it even survived malfunctioning hardware, without data corruption (only one consistency error which could be corrected automatically).
Anyway, Reiser4 should be worth trying. As far as I know it is a complete rewrite, so stability concerns of the past do not apply anymore.
I encounter a crash with reiserfs last week (where home reside 🙁 ) and no reiserfs tool let me recover and boot safely. With knoppix I was able to save nearly everything.
After reinstalling suse 9.0 I remark that reiserfs default setting for journal is “Ordered” and not “journal”. THis can explain why the change are not journalized and why a sudden power off can kill a filesystem.
Otherwise, i confirm running 2000 or XP and a sudden power off while doing defragmentation OR scaning with anti virus kill the disk (I test both case recently, I am using pc since 1993 and never had so much bad luck)
defragmentation + poweroff -> no filesystem on disk, NDDoctor, scandisk, no efffect with SUse I was able to save all data.
antivirus + poweroff -> blue screen when 2000 sp5 start !!!!
There is no way you could lose a NTFS partition only because of an antivirus scan.
>> RaiserFS bites on power outage.
I don’t think so. I have tried it many times, even with the button (yes, the one that has RESET printed above it)
I have used reiserfs since the 2.2.x days (patched, of course), and have used it ever since.
one thing that the people criticizing reiserfs must keep in mind is the the hash that they’re going to make the filesystem with (r5, rusparov…). read some documentation, and you will see that some trade speed for reliability, and it would be interesting to see what hash the people complaining about reiserfs (and the ones praising it) used when formating their partitions with reiserfs.
I don’t have anything against the other filesystems, as I have used all of them, but I choose reiserfs because I NEVER lost data yet, and since I don’t have a UPS for my PC, xfs and jfs are out of question for me…
I tried JFS once. As a Swede I like to use the letters å, ä and ö in some of the names of files and directories. However, these letters showed up as “?” when using both ls and Nautilus The files and directories where all “untouchable” i.e. I could not rm, cd, mv or cp them and ls returned error messages when it ran in the parent directory. After reformatting the partion to ReiserFS the problem went away immediately.
My story for the great book “Journaling Filesystems in Practice”.
I once had a problem related to XFS. My desktop system crashed somehow (no power outage). After i rebooted, mount couldn’t do its work. I tried to boot using some old CD or floppy (can’t remember) with XFS support but that didn’t work either. I thought i was fried. I really thought i lost my data. At a desperate moment i went to Freenode’s #xfs (where the developers & community reside) and someone from there suggested to download & burn a CD with support for Linux XFS’s new version, then mounting the system. So i tossed in a CD, mount, and unmount the filesystem. That worked. What was the problem? Most likely the kernel didn’t understood the journaling log. After this i recompiled a new kernel with support for the newest Linux XFS version. Putted that on my system. Worked like a charm.
This sounds a bit like the “I encounter a crash with reiserfs last week (where home reside 🙁 ) and no reiserfs tool let me recover and boot safely. With knoppix I was able to save nearly everything.” problem. And seriously, was my system really fried? No it wasn’t. It was plain ignorance from my side. Makes me wondering how many of the occured problems mentioned here were actually justified as a technological faillure. Who knows? Maybe it was the users’ fault, or it was the user who didn’t know how to solve it while it was solable as in my case.
Also i had a power outage 2 weeks ago. All my computers running Ext3FS were still running fine. I’ve got no missing files. Which is <doh> with Ext3FS conservative default options (see man tune2fs and then the -o option).
I regulary just put my SGI machines using XFS off the hard way; no problem. Heh, and they should, given they’re normally used for Really Important purposes.
I’m btw interested in performance of cryptoloop/dmcrypt/similar FSes. Anyone else interested in that?
http://slashdot.org/comments.pl?sid=107168&cid=9116522
In this example the user found out the reason why ReiserFS got corrupted and data got lost.