New filesystem: Btrfs!

Message

ToeiRei · Post by **ToeiRei** » Thu Aug 09, 2007 7:31 am

A word of warning:

The new defrag thing on btrfs 0.6 causes a kernel oops here - or at least that's what it looks like. I have already notified the dev. mailing list.
V 0.5 works fine here.
Could someone please confirm this bug?

Rei

ToeiRei · Post by **ToeiRei** » Fri Aug 10, 2007 9:19 pm

Ok folks, the btrfs kernel module 0.7 is out;
the kernel oops bug looks fixed.

Btrfs-devel mailing list wrote:Hello everyone,

The defrag code in v0.6 seems to have been really good at triggering a
bug in my extent tree allocation code. So, I've put out v0.7 with the
fix. There are also a bunch of defrag refinements and Josef's fsync
improvments included.
...

Greetings and happy testing

Rei

kernelOfTruth · Post by **kernelOfTruth** » Wed Aug 15, 2007 5:27 pm

nice !

GNU/Linux definitely needs a competitive filesystem against ZFS:

http://www.sun.com/software/whitepapers ... cid=920608

ToeiRei · Post by **ToeiRei** » Wed Aug 15, 2007 5:41 pm

I am doing as much testing as possible to give feedback to the developers.
Didn't have any crashes so far with 0.7 by now. And it's a great filesystem for portage

Roman_Gruber · Post by **Roman_Gruber** » Wed Aug 15, 2007 5:52 pm

kernelOfTruth wrote:nice !

GNU/Linux definitely needs a competitive filesystem against ZFS:

http://www.sun.com/software/whitepapers ... cid=920608

thx, I read that.

arnvidr · Post by **arnvidr** » Thu Aug 16, 2007 7:00 am

Cool. I find out about a promising new file system and get some nice quote for my signature on the same day from the same thread!

I'll be following this one for sure!

zxy · Post by **zxy** » Fri Aug 17, 2007 11:40 pm

Are there any new benchmark/tests done on btrfs?

I'm especially interested in comparison against reiser4.

kernelOfTruth · Post by **kernelOfTruth** » Sat Aug 18, 2007 12:56 am

zxy wrote:Are there any new benchmark/tests done on btrfs?

I'm especially interested in comparison against reiser4.

DIY ?

zxy · Post by **zxy** » Sat Aug 18, 2007 12:59 am

kernelOfTruth wrote:
zxy wrote:Are there any new benchmark/tests done on btrfs?

I'm especially interested in comparison against reiser4.
DIY ?

you might be right

ToeiRei · Post by **ToeiRei** » Wed Aug 29, 2007 12:55 pm

I will do some more benchmarking at a later time as there are some big changes being implemented.

Rei

ToeiRei · Post by **ToeiRei** » Wed Aug 29, 2007 1:50 pm

I have placed /var/tmp/ccache on my btrfs volume - here are some stats. Cold ccache (ccache -C, then emerge...)
I will try to do the same tests with some bigger apps.

Code: Select all

 * sys-apps/sed

     Tue Aug 14 15:34:25 2007 >>> sys-apps/sed-4.1.5
       merge time: 25 seconds.

     Wed Aug 29 15:45:47 2007 >>> sys-apps/sed-4.1.5
       merge time: 22 seconds.

Code: Select all

 * dev-libs/expat

     Tue Aug 14 15:34:00 2007 >>> dev-libs/expat-2.0.1
       merge time: 24 seconds.

     Wed Aug 29 15:48:26 2007 >>> dev-libs/expat-2.0.1
       merge time: 18 seconds.

kernelOfTruth · Post by **kernelOfTruth** » Wed Aug 29, 2007 2:47 pm

try throwing your portage tarball at that partition (after having compressed it), don't use the tarball from the net since it seems to lack the licenses & therefore lots of small files

alternatively copy loads of random files to it (e.g. your home if there's much different files)

could you make some tests with reiser4, reiserfs, btrfs, ext4 if available & possible ?

last time I tried it - it was already faster than reiser4

now it should be even faster

I'm waiting for the compression thingy - then that filesystem will outdo every filesystem on the market

ToeiRei · Post by **ToeiRei** » Mon Sep 17, 2007 10:54 am

Ok guys: btrfs 0.8 and btrfs-progs 0.8 are out - grab them as long as they're hot

Most important change for me is the subvolume mounting which can be done directly by now:

Code: Select all

mount -t btrfs -o subvol=foo /dev/whatever /mount/point

kernelOfTruth · Post by **kernelOfTruth** » Thu Sep 20, 2007 8:33 pm

nice one

here another test concerning space efficiency:

a portage-partition + some additional packages on it:

/dev/sdc7 reiserfs 4,7G 318M 4,4G 7% /usr/portage
/dev/sdc7 btrfs 4,7G 348M 4,4G 8% /test

ls -l /usr/portage/packages/All/
total 63609
-rw------- 1 root root 2623679 2007-09-16 00:57 bmpx-0.40.3.tbz2
-rw------- 1 root root 41592229 2007-08-18 21:51 gcc-4.2.1.tbz2
-rw------- 1 root root 51868 2007-08-18 19:13 gcc-config-1.3.16.tbz2
-rw------- 1 root root 1407158 2007-08-22 03:54 gettext-0.16.1-r1.tbz2
-rw------- 1 root root 256768 2007-08-30 22:42 glib-1.2.10-r5.tbz2
-rw------- 1 root root 18030807 2007-08-21 17:21 kdelibs-3.5.7-r10.tbz2
-rw------- 1 root root 1094366 2007-08-18 22:57 pdftk-1.41.tbz2

so it seemingly got better at that

enderandrew · Post by **enderandrew** » Fri Sep 21, 2007 2:41 am

Is that stock reiser4 or reiser4 with compression?

And I'm about to wipe my root partition and start over. I'm likely going to go back to reiser4, but this looks pretty sexy. However, the amount of corruption reported early in this thread is scaring me.

kernelOfTruth · Post by **kernelOfTruth** » Fri Sep 21, 2007 9:14 am

you refer to my last post above yours ?

it's actually reiserfs v3.6 I'm using since reiser4 isn't compiling on >=2.6.23-rc7,

btrfs is also faster than reiser4, reiser4 with cryptcompress lzo1-compression is still the fastest filesystem

you still can't patch this filesystem into the kernel & only use it as a module - are you serious you want to use it for root ??

ejs76 · Post by **ejs76** » Fri Sep 21, 2007 9:25 am

@enderandrew

From the announcement of this new filesystem on the kernel mailing list (http://kerneltrap.org/node/8376)

The current status is a very early alpha state, and the kernel code
weighs in at a sparsely commented 10,547 lines. I'm releasing now in
hopes of finding people interested in testing, benchmarking,
documenting, and contributing to the code.

I've gotten this far pretty quickly, and plan on continuing to knock off
the features as fast as I can. Hopefully I'll manage a release every
few weeks or so. The disk format will probably change in some major way
every couple of releases.

This says enough I guess.

enderandrew · Post by **enderandrew** » Fri Sep 21, 2007 9:39 am

kernelOfTruth wrote:you refer to my last post above yours ?

it's actually reiserfs v3.6 I'm using since reiser4 isn't compiling on >=2.6.23-rc7,

btrfs is also faster than reiser4, reiser4 with cryptcompress lzo1-compression is still the fastest filesystem

you still can't patch this filesystem into the kernel & only use it as a module - are you serious you want to use it for root ??

I'm blind, as the post clearly says reiserfs, and not reiser4.

No, I don't think I'd risk using this for root. Still probably going reiser4/lzo

Matteo Azzali · Post by **Matteo Azzali** » Sun Sep 23, 2007 10:17 am

kernelOfTruth wrote: btrfs is also faster than reiser4, reiser4 with cryptcompress lzo1-compression is still the fastest filesystem

Hum, where you got that results? Which test case? I tested it with iozone (version 0.6, maybe) some
weeks ago but I haven't noticed it was so fast , maybe I played too much snapshotting.....

alexandervdm · Post by **alexandervdm** » Tue Oct 02, 2007 9:48 am

arnvidr wrote:Cool. I find out about a promising new file system and get some nice quote for my signature on the same day from the same thread!

I do hope people realise that I was being sarcastic, because your signature makes me look like an idiot

arnvidr · Post by **arnvidr** » Tue Oct 02, 2007 11:07 am

AforAlexander wrote:
arnvidr wrote:Cool. I find out about a promising new file system and get some nice quote for my signature on the same day from the same thread!
I do hope people realise that I was being sarcastic, because your signature makes me look like an idiot

Well, I was away from this forum for two years, you would know about the general intelligence of the people here more than me I guess. If you want I'll add something indicating it was sarcastic

Or did you mean that people could connect it with the other quote in my signature? I could swtich them around perhaps...

steveL · Post by **steveL** » Tue Oct 02, 2007 12:18 pm

arnvidr wrote:
AforAlexander wrote:I do hope people realise that I was being sarcastic, because your signature makes me look like an idiot :P
Well, I was away from this forum for two years, you would know about the general intelligence of the people here more than me I guess. If you want I'll add something indicating it was sarcastic ;)
Or did you mean that people could connect it with the other quote in my signature? I could swtich them around perhaps...

AforAlexander wrote:Good thing data corruption is only a minor issue in filesystems.

AidanJT wrote:Do you make a habit of being completely and utterly wrong?

Ouch! That's bad ;P I'd definitely switch em around (and maybe add a <sarcasm> thing.)

Matteo Azzali · Post by **Matteo Azzali** » Thu Oct 04, 2007 1:32 pm

AforAlexander wrote:Good thing data corruption is only a minor issue in filesystems.

As this is gonna be misinterpretated more and more, let's clarify some little thing, right?

1) While it's not really clear the meaning AforAlexander was givin to this phrase, using it alone is
"decontexting it", and gives a different meaning. This beacause:
-------a) the discussion was about an Alpha released filesystem (for which data corruption is really
----------a minor issue as it will get fixed for sure before beta releases)
-------b) the following phrase,

J/k, keep up the good work

-----------means it wasn't malicious or so.

2) As still most "different than ext3" filesystems seem to be hatred by some users,
Let's clarify once for all what the ext2/3 lesser data corruption is:
when experiencing a crash or a blackout, other fs are likely to not having saved your last saved file
(if it was saved just before the crash/blackout), while ext2/3 has more chanches to have this done better.
in non-ext3 filesystem there is no corruption of anything that wasn't about to being written (and I repeat, written)
when the crash/blackout happens
(yes, this is vital in some server applications but's really not needed for most users, expecially if the drawback is
a 4,5 minutes fsck on a 70Gb partition, "suggested" once every 24 boots (that's 270 seconds, so 11,25 seconds more
for each boot, splitting)).
And that's why I'm regretting my ext3 choiche, shame on who suggested it to me, the only real advantage a desktop user
may find in ext3 is for "/var/log": a line more in a logfile may mean the cause of a crash.

As for the sign, arnvidr just read this sign and say what would you think about it:

arnvidr wrote:Besides, I think the name HD-DVD isn't winning any votes. It just seems like a lazy upgrade (which is probably why Microsoft likes it, that's their business model after all) and BluRay feels more "hip". Silly - yes; working - yes.

AidanJT wrote:Do you make a habit of being completely and utterly wrong?

....understood what's the issue?

StifflerStealth · Post by **StifflerStealth** » Fri Oct 05, 2007 12:39 am

kernelOfTruth wrote:are you serious you want to use it for root ??

I am crazy enough to make it my root fs.

I would call it "fun".

Anyways, I definitely will try this fs out. I am getting tired of plain old, ordinary ext3. Reiser4 was never stable enough to keep a system one, but that was beta and I have been too lazy to redo my system. I can't seem to tarball the system and format. Break out the boot discs that support this system.

I agree with everything Matteo Azzali said.

I couldn't have said it better myself, if I would have said it, but I didn't say it, so ...

EDIT: Here's a link to a presentation that was done. It explains a lot and has benchmarks.
http://oss.oracle.com/projects/btrfs/di ... -ukuug.pdf

ToeiRei · Post by **ToeiRei** » Fri Oct 05, 2007 6:07 am

I second that - crazy enough too.