Filesystems comparison for present time (r4,r3,jfs,xfs,ext3)

[fallow]

Hey.

Yeah . Propably many of You like reiser4 and etc. This thread is not to discuss who are liking it and the reason of liking it or not liking it Im a user of filesystem . I did a simple test to compare a simple things which I doing day by day on my box.

Why ?
I was used reiser4 for a long time. And I was not happy because of some things. I Switched to reiser3. Im not 100% happy because of high cpu usage with it. (dont ask me for reiser4 . In the past I used ext2 and ext3 with the redhat for a long time, also jfs and xfs on Gentoo. JFS code has some improvements and I wanted to see what is the current situation at present time.

so . this is a results .

my box : AMD 1.0 Athlon , 256 MB Ram
HDD used for tests IBM
hdd: IC35L040AVVA07-0, ATA DISK drive
hdd: max request size: 128KiB
hdd: 80418240 sectors (41174 MB) w/1863KiB Cache, CHS=65535/16/63, UDMA(100)
all tests with the same kernel , IngoSched ,CFQ-TS IO Scheduler , and latest versions of all fs code.

1. FIRST - bonnie++ test (remember that us is smaller than ms , so better in latency )

[Enlight]

Well currently I'm an xfs user, and I was thinking that it was time using it a better way, it'seems that there are somme great features available like putting journalisation on another hd than the one you're writting on wich should massively improve latency (I was thinking about getting an old hd for stuffs like this) and I've heard that there were ways to makes it perform very fast on small files.

But now your post make me want to give a try at JFS, but I was wondering if multiples files sytem support won't perform slower (ie using reiser3.6 for /etc, xfs for /home...) for 1) making kernel bigger and 2) increasing cpu load?

[Cintra]

Hei
I currently use reiser3.6 but I'm thinking of putting up a test 2005 partition using jfs
the following post helped make up my mind..

[fallow]

Hmm. Yeah . journal data on other hd can be a very good idea. but I dont have possibilities to do it

my target is a home desktop

In the past I had many of different fs for /usr , portage etc. but now I want to stick with the one.

I think that ext3 can be a best choice.

cheers.
_________________
"Time is a companion that goes with us on a journey. It reminds us to cherish each moment, because it will never come again. What we leave behind is not as important as how we have lived" J-L. Picard

[IvanYosifov]

fallow, it may be my browser fonts, but the text in the table is missaligned Can you please put the data in a plain text file an upload it somewhere ?

[Enlight]

Yes I think about sticking with only one too, in order to have an as small as possible kernel, that's why I was wondering If stuffs like initrd could be done with xfs as an example, putting journalisation elsewhere.

BTW I was thinking for desktop too but as I 've got a sata disk I was thinking about getting an old and cheap little ide disk to put journalisation on it, or why not an usb key... or something insane like this

ps : I'm currently trying vivids which seemed to be the closest sources from what I was looking for, and would soon report back my impressions on 'em

[vipernicus]

For a while, I had decided to switch to ext3 from reiser3, and found a major issue, at times, when moving or copying a bunch of < 25kb files, the file system would lock up, and cause me to reboot, this has actually happened to me on more than one occassion, with gentoo, and with slackware. With reiser3, i never have this issue. I am curious about JFS though.

[Sith_Happens]

Here is a pretty good benchmark I found a while ago for xfs, jfs, xext3, ext2, and reiserfs 3.6. Before you say the benchmark is too old, look at the hardware they are using. I think your benchmarks are good fallow, just check these out too. Anyway I use Reiserfs 3.6, I think it has the best mix of stability and speed for my application. However, if you look at the benchmark, you'll see that each filesystem has it's advantages and disadvantages, it's not really like there is a "best filesystem" for all purposes. However, it does seem that this benchmark shows there is worst filesystem for all applications ext3. We'll, not for all aplications, it is stable as hell, but it's also as slow as it is stable.
_________________
"That question was less stupid; though you asked it in a profoundly stupid way."
I'm the brains behind Jackass! | Tutorials: Shorewall

[Enlight]

[Sith_Happens]

[ewan.paton]

i uses xfs for /home reiserfs for / and may get round to switching /usr/portage to reiser4 eventualy i cant quite place why i like xfs for big files but i do, i think its about matching whats best for the job
_________________
Giay tay nam | Giay nam cao cap | Giay luoi

[Illissius]

(You did reboot or otherwise clear the filecache between tests, right?)

Thanks a lot for the numbers. For JFS's and XFS's low performance and CPU utilization during the file create/remove tests, the explanation is right below: incredibly high latency. This is not a good thing, it means system resources can't being fully utilized. EXT3 also has very high latency (though not as high as X/JFS), but that apparently doesn't prevent it from consuming 100% CPU while performing piss poorly. These numbers also explain why reiser* is so fast at emerge syncing, while not necessarily at other things: they're very fast at file creation/deletion, which emerge sync apparently involves lots of. With the exception of reiser4 rather underperforming, the data read/write tests don't seem to have much variance, which makes sense, as I would assume there's less filesystem involvement anyways (filesystems manage the stuff between the files, not so much the stuff in them).
I'm also surprised about the kernel source and livecd copy/remove test results... the traditionally expected results would have been reiser* dominating the copy/remove tests, and XFS likewise for the livecd copying, which plainly did not happen.

Anyways, the conclusions I would draw from this data (the results would have to be independently verified before I'd put any trust in it):
- reiser4 doesn't (yet) live up to the hype. at best it manages to keep pace with the others, at worst it's significantly off it. coupled with the potential stability issues, it doesn't look too attractive.
- XFS and JFS suck at both file creation/deletion and copying large files, while not making up for it in the other tests. at least they are likely to actually *work*, but nothing to get excited about. (such high latency and poor performance for XFS in copying a large file make me think that there may have been issues with the test configuration in fact, hence the disclaimer above...)
- EXT3 is fucking abysmal at creating/removing files, but performs very solidly at everything else. Assuming you don't mind your emerge syncs going slower (is there anything else that involves heavy file creation/removal besides that?), and given its solid track record, it's a fine enough choice.
- reiser3 leaves everything else in its dust for creating and removing files (only reiser4 gets close), and keeps up quite well in all the other tests. and has a likewise solid track record. it would, hence, be my pick.
(Again, let me reiterate that these conclusions are based on, and only on, the data presented above. More data (such as XFS actually not sucking at the thing it was meant to be good at) may well end up changing them.)

And thanks again to fallow for making the data available.
_________________
Work is punishment for failing to procrastinate effectively.
last.fm

[Lechium]

This is your choice, really... I've used all of these aside JFS, and honestly couldn't tell the difference between ext2 and reiserfs bu simply using these. Tests and such show the descrepency, but as end user experience goes the differences are miniscule.
On the side note -- pick whatever you like as long as it has journaling. Journaling will slow things down (by a hardly noticeble amount), but wil avoid possible half-writen files etc, so IHMO its worth it.

P.S. reiserfs4 can be the best choice for a simple reason -- a lot of people use it, so if you get into trouble it will be easier finding help, than if you run jfs/xfs.

[codergeek42]

[i92guboj]

[Lechium]

[fallow]

Im little bus at the moment , and not reading all but

is the one thing - > speed of filesystem isnt all . In most cases faster fs occupied this by lost of interactivity . Is important to rethinnk what do U want - faster FS at all costs , or good overall interactivity . Im trying to find a good balance beetwen everything now .

and I think that is must to have better testing than this . this is _very_ very simple test.
JUST to start a discussion . Is many of other better methods . (OH sorry bonnie++ results was good )

[b] ANOTHER TEST [b]
One of friends from Polish forums did testing based on the good this ( http://linuxgazette.net/102/piszcz.html ) test.

[codergeek42]

[Illissius]

[i92guboj]

This is the a table that resumes the one that fellow posted above, here we can see the thing in a compact way.

[Kraymer]

The mix is important.
Currently I use reiserfs (3.6) for root and home, xfs for my data partition (file server, lots of bigg files).
Once it's ready (whenever it'll be) I'll change my system and home to reiser4.

Had bad experience with jfs once: On my fileserver I had major data loss on power failure, the partition was unmounable (and believe I searched my ass off the Inet and it turned out that I was doomed (special circumstances though). Switched that to xfs, and even on a software raid, power failure happend again, system rebooted fine and everything was in its place

You always have to consider the purpose the filesystem shall fulfill. For overall usage, reiserfs is fine for me (although it's sad that only 8K blocksize works ), xfs is great for my large data partition.

Sebastian

[spb]

[Enlight]

okay so I guess I'll definitively stick with xfs, morover I tried reiserfs 3.6 once and my sata disk was making twice noise as in xfs while writing... gonna got a usb stick and try to put journalisation on it

[Kraymer]

[Enlight]