Linux.com

Since Hans Reiser was recently <a href="http://sfgate.com/cgi-bin/article.cgi?f=/c/a/2006/10/10/BAGERLM3RR15.DTL" title="sfgate.com">arrested in the case of his wife's disappearance</a sfgate.com>, it sounds like it may not be a bad idea.

But, why use ext3? Why don't they use Novell's NSS(Novell Storage Services)? It's fast, reliable, allows block level access for backups etc, has built in deleted file recovery, and lots more.

#

...SUSE users can still have Reiser. All they did was make ext3 the default.

#

What about XFS?

#

Uhm...you can use that FS too if you like.<nobr> <wbr></nobr>;-) This really is no big deal, only the default FS has been changed, nothing else.

#

I've always used ReiserFS . At first I read it's something cool and less conservative than the good old ext3 . That's why I'm going to continue using it because I like living on the edge and ext3 is just too dull , I might use it for a server but never for my desktop.
ReiserFS never let me down and I've never had problems though to be honest I never did test its limits and only used it for simple PC duty.
If I'd have to switch from Reiser though I'd go for XFS for sure . ZFS from Sun is also nice but won't be in Linux ( in the kernel ) for a while ( if ever ).

#

Perhaps only the implementation is copyrighted under CDDL.
But maybe possible for people to look at documentation and write a ZFS compatible clone.
Example, Linux can use NTFS, but NTFS is copyright by Microsoft, but they made a clone.

#

===========
PERFORMANCE
===========

Our highest priority is reliable functionality, performance comes only
next. However acceptable performance is part of the functionality so we
constantly keep an eye on it and consider any major performance issue to
be a real functionality problem as well.

Below are the averaged file creation, deletion and access performance
numbers got by running 'bonnie++ -s0' using several filesystems (the
benchmark used 16,000 files per directory in each sessions):

Version  1.03    ------Sequential Create------ --------Random Create--------
                -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
          files <nobr> <wbr></nobr>/sec %CP <nobr> <wbr></nobr>/sec %CP <nobr> <wbr></nobr>/sec %CP <nobr> <wbr></nobr>/sec %CP <nobr> <wbr></nobr>/sec %CP <nobr> <wbr></nobr>/sec %CP
reiserfs-----16k 21459 99 +++++ +++ 17856  96 20172  98 +++++ +++ 16414  96
jfs----------16k  7015 13 +++++ +++  5868  10  3068  14 +++++ +++  1075  3
ntfs-3g------16k  3021 99 14291  99  5226  99  3548  99 16149  99  5223  99
xfs----------16k  2401 17 +++++ +++  2095  15  2301  20 +++++ +++  347  2
ext3---------16k  1862 96 +++++ +++ +++++ +++  1914  96 +++++ +++  9695  98
minix--------16k  1450 97 +++++ +++ 18148  94  1694  97 +++++ +++  4847  98
fat32--------16k   366 97 +++++ +++  1809  97  428  97 +++++ +++  1361  97
paragon ntfs-16k    58 98  1259  99  245  99    55  99 +++++ +++  832  99
captive ntfs-    Always crashes early on file creations and loses files-
                This was also confirmed and fix denied by its developer.

We estimate the ntfs-3g driver performance could be increased by several times.

Well ReiserFS seems HUGELY better than ext3,... so no wonder Novelle is swapping to ext3.

Jade @ <a href="http://linux.coconia.net/" title="coconia.net">http://linux.coconia.net/</a coconia.net>
Jade @ <a href="http://m.domaindlx.com/LinuxHelp/" title="domaindlx.com">http://m.domaindlx.com/LinuxHelp/</a domaindlx.com>

#

Somehow, Jade, your posts are extremely recognizable, even when you forget to put your off-topic plug for your own personal soapbox, coconia.net, at the end. Please, please, please, for all our sakes, develop a less annoying style of posting, stop blathering like the bombastic fool that you are, and STOP TOP POSTING WHEN YOU MEAN TO REPLY!

And yes, the caps do mean I am yelling. I thought I'd clarify since you seem to have been significantly absent when the senses of etiquette were handed out.

#

The above table is a little hard to read, so here are the relevant figures:

Sequential Create
files/sec %CP
21459 99 (ReiserFS Create)
 1862 96 (ext3 Create)

Random Create
files/sec %CP
20172 98 (ReiserFS Create)
 1914 96 (ext3 Create)

Random Delete
16414 96 (ReiserFS Delete)
 9695 98 (ext3 Delete)

Jade @ <a href="http://linux.coconia.net/" title="coconia.net">http://linux.coconia.net/</a coconia.net>

#

Oh, I forgot this one:

Sequential Delete
17,856 96 (ReiserFS Delete)
+++++ +++ 98 (ext3 Delete)

The +++++ +++ actually means that ext3 is so fast at sequential deletion that an accurate result cannot be given as rounding errors in measurements become the determining factor.

The only problem is, that significant sequential deletions never occur in real life computing.

Jade @ <a href="http://linux.coconia.net/" title="coconia.net">http://linux.coconia.net/</a coconia.net>
Jade @ <a href="http://m.domaindlx.com/LinuxHelp/" title="domaindlx.com">http://m.domaindlx.com/LinuxHelp/</a domaindlx.com>

#

Good job, Jade! You used the reply button like a good boy. Here's some candy.

Perhaps I'm misunderstanding what is meant by sequential deletions... I have many many times deleted an entire directory tree, with many hundreds, sometimes even thousands, of files. It would seem that when this is done, all of those files are deleted<nobr> <wbr></nobr>... sequentially!

#

"Perhaps I'm misunderstanding what is meant by sequential deletions,..."

True,... Seems you don't know what you are talking about.

All deletions would be sequential by your definition.

Sequential deletions are deletions of files written sequentially on the hard drive.

Random deletions are deletions of files written to random places on the hard drive.

#

Going from R3 to R4 would require a reformat. To avoid that, they're going from R3 to ext3, which<nobr> <wbr></nobr>... requires a reformat if you're already using R3.

I thought one of Mahoney's comments was less than candid - to the point of being dishonest:

Ext3's performance in some situations may not be on par with Reiser3,

Yeah, right. Like a factor of 10 worse on file creation, and nearly twice as slow on file deletion. I really hate waiting for my computer. Why do so many developers seem to think performance is unimportant?

#

First, as a poster said earlier, it's only a change in the default for new installations. Nobody is being deprived of R3.

That said, leaving R3 for another filesystem is guaranteed to require a re-format. But what about the next filesystem upgrade after that? How about ext3 to ext4? We already have the ext4 developers' word that it will be a clean, non-destructive upgrade, just as ext2-to-ext3 was. Will R5 require a re-format to upgrade from R4? We don't know yet. Given the re-format required for R3-to-R4, it would be unwise to expect a non-destructive upgrade from R4 to R5.

Now, for those "performance" figures...

Seriously, what program or user creates 16,000 empty files all at once? A program may create 5, 10, 20, or even 100 temporary files, but those are usually held in<nobr> <wbr></nobr>/tmp, mounted as a tmpfs so that it uses no actual disk access other than swapping. Even if they aren't, (1) filesystem caching will help the performance, but (2) are you really going to notice if The GIMP takes an extra half-second to launch?

Actually, odds are that launching The GIMP on an ext3-based SuSE system will be faster. For now, at least, ext3 on SuSE means it's a new system, with little to no filesystem fragmentation.

The big exception to this is unpacking large source tarballs, like the Linux kernel (25,225 files in my Slackware source tarball). Even then, is your kernel build so urgent that 15 seconds more to unpack is a matter of life or death? And how many times a day are you unpacking source tarballs, anyway? 5, 10, 20, 100?

(I'm ignoring the build process. It's CPU-bound, no matter the filesystem.)

Don't blindly quote benchmarks. Think instead about SuSE's target audience, and real-world usage.

#

Gus3 says: Don't blindly quote benchmarks.

The turkey Gus3 would rather quote NO benchmarks at all.

Or, lie about the benchmarks that are considered.

The -s0 option suppresses the Sequential Output/Input part of the test.

Here is the test carried out on an old box (hence figures are lower than those of Szaka - only the relative performance counts) with the -s option - with file-size 2016M - the file-size for the second part of the test is 16K as before:

Gus3 says: Don't blindly quote benchmarks.

______------Sequential Output------ --Sequential Input- --Random-
______-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--

  Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP__/sec %CP
2016M 38093__61 42896__27 15334___7 31245__38 34748___9 166.3___0
______------Sequential Create------ --------Random Create--------
______-Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files__/sec %CP__/sec %CP__/sec %CP__/sec %CP__/sec %CP__/sec %CP
___16 13683 100 +++++ +++ 14125__93 13272__99 +++++ +++ 12984__93

#

"the file-size for the second part of the test is 16K as before:"

Should have been:

the number of files for the second part of the test is 16K as before:

These are small files in a fashion similar to the usage patterns of programs such as Squid or INN (according to the docs).

#

Sir, I have said it to you before, and I am now repeating myself. LEARN TO USE THE REPLY LINK! Top posting for the purpose of a reply is EXTREMELY RUDE.

#

The default for this test (the tests reported above) is to test with 0 byte files.

To use files of other sizes you can specify -n number:max:min:num-directories

My apologies.

#

There are some comparisons here:

<a href="http://www.fortunecity.com/skyscraper/romrow/935/jfs_xfs_rfs_ext.html" title="fortunecity.com">http://www.fortunecity.com/skyscraper/romrow/935/<nobr>j<wbr></nobr> fs_xfs_rfs_ext.html</a fortunecity.com>

I don't know how to read them.

#

I'm a ReiserFS user as well, and dislike the switch of 'default' to ext3. I lay most of this at the feet of Mr. Reiser. His attitude towards improving 3.6 and providing a clean upgrade path from 3.6 is probably as much a cause of this as anything. He has also ruffled quite a few feathers with Reiser4, in particular with the kernel developers. I suspect any fair minded person would conclude he is being treated like a red-headed stepchild by those same developers. That same person would also find that a good bit of blame should be laid on his doorstep as well. I find it particularly galling that ext4 has been accepted to the development kernel, while Hans has been fighting for over a year to get Reiser4 into the kernel. I know it's inevitable, but I hate to see the spectre of politics and buddy networks raising their ugly head in what should be a purely technical consideration. Hans is brilliant, but like many brilliant people his drive for excellence can come off as arrogance and antagonism.

#

I'm a ReiserFS user as well, and dislike the switch of 'default' to ext3. The only reason I see the switch to ext3 is that it is much more widely supported than reiserfs, and has an upgrade path. I only wonder what kind of filesystem Reiser 3/4 would be if 1/10th the support and work that has been expended on ext3 was spent instead on reiserfs's.

I lay most of this at the feet of Mr. Reiser. His attitude towards improving 3.6 and providing a clean upgrade path from 3.6 is probably as much a cause of this as anything. He has also ruffled quite a few feathers with Reiser4, in particular with the kernel developers. I suspect any fair minded person would conclude he is being treated like a red-headed stepchild by those same developers. That same person would also find that a good bit of blame should be laid on his doorstep as well. I find it particularly galling that ext4 has been accepted to the development kernel, while Hans has been fighting for over a year to get Reiser4 into the kernel. I know it's inevitable, but I hate to see the spectre of politics and buddy networks raising their ugly head in what should be a purely technical consideration. Hans is brilliant, but like many brilliant people his drive for technical excellence seems to be seen as arrogance.

#

Um, he is arrogant and antagonistic. On the LKML lists he consistently refuses to answer technical questions or discuss legitimate concerns, but instead avoids them and falls back on insults and flames.

ReiserFS does not scale and it is not reliable. It does well as long as it's not stressed, but under stress it falls apart. We've been testing Reiser, XFS, JFS, and Ext3 for a couple of years, and all the longevity tests favor Ext3. All the speed and shiny new features in the world don't matter if the filesystem is fragile, and I'm afraid that Reiser is too fragile for heavy-duty use. Mr. Reiser does not help by being obstructive and denying that problems exist.

#

Damn right it's not reliable. If you have to pray every fucking time you reboot to see if your files are still there, it's not worth it to waste time with such a lame FS. Good riddance!

#

You are talking crap.

#

No, you are. Now shut that ass of yours you call a mouth.

#

Nothing stupider than a name-calling flame war between two juvenile anonymoids.

#

"We've been testing Reiser, XFS, JFS, and Ext3 for a couple of years"

No kidding; show us the results.

Publish them here.

#

"This Ext3 has a history of not requiring reformats to accomplish an upgrade,..."

IS TOTAL BULLS**T.

The reason you don't need to reformat is because Ext3 is Ext2 with journeling tacked on.

You had to reformat to go from ext1 to ext2, and you will have to reformat to go from ext3 to ext4 (unless ext4 has no interesting new features).

And anyway, who cares if one has to reformat. I have NEVER upgraded a file-system without reformatting. Reformatting is no big deal.

#

If you ask me, many kernel developers seem to be the "ugly" people in all this.

Hans Reiser, seems to be just reacting to a totally unjust conspiracy against his product.

#

Yeah, well until Hans can respond maturely and accurately address those concerns about his FS (which ARE legitimate), he has no business in telling people what to include in the kernel or not. There, I said it.

#

Prove to me that the concerns about his FS (Reiserfs) are any more legitimate than corresponding concerns about the ext4 FS.

Yet ext4, with ALL it faults, is in the kernel.

#

with all respect, ext4 is an incremental upgrade from ext3. They separated it in the first place to guarantee continuous stability for the users.

I can agree however that developing filesystems shouldn't be done in stable branch of the kernel in the first place, neither extX nor reiserfsX.

#

Real Life File System Tests.

cp<nobr> <wbr></nobr>/linux-2.4.17.tar.gz<nobr> <wbr></nobr>/mnt/fs
(cd<nobr> <wbr></nobr>/mnt/fs && tar xfz<nobr> <wbr></nobr>/mnt/fs/linux-2.4.17.tar.gz)
rm -fr<nobr> <wbr></nobr>/mnt/fs/linux-2.4.17.tar.gz<nobr> <wbr></nobr>/mnt/fs/linux

___________ jfs _____ xfs ____reiserfs__ reiserfs____ ext3______ ext3______ ext3______ ext2___
___________ jfs _____ xfs ____reiserfs__ (notail)__(writeback)_(ordered)__(journal)___ ext2___
real 1m_23.683s 0m_36.143s 0m_35.545s 0m_19.582s 0m_25.984s 0m_25.152s 0m_34.794s 0m_21.498s
user 0m__4.930s 0m__4.890s 0m__5.080s 0m__4.900s 0m__5.110s 0m__4.860s 0m__5.030s 0m__4.580s
sys_ 0m__3.810s 0m__8.570s 0m__6.410s 0m__5.590s 0m__3.150s 0m__3.580s 0m__3.400s 0m__2.640s

reiserfs (with notail) wins EASILY (among journaling filesystems (the non-journaling filesystem, ext2, is close).

January 12, 2002 © Sergey I. Rotar

<a href="http://www.fortunecity.com/skyscraper/romrow/935/memxfer5b.html" title="fortunecity.com">http://www.fortunecity.com/skyscraper/romrow/935/<nobr>m<wbr></nobr> emxfer5b.html</a fortunecity.com>

#

Perhaps you have made a mistake. The copyright date on the page you quote here is in fact "Copyright (c) by Sergey Rotar, Jun 17 2001". I'm surprised, what with your adeptness at the copy and paste, that you managed to screw that one up. Perhaps you were trying to make it appear that these results were not quite as far in the past, and thus marginally more credible and useful? If so, I am ashamed of you.

#

Apologies; Gave the wrong link (you didn't even notice,.. tut, tut).

Have a look at the date on this page:

<a href="http://www.fortunecity.com/skyscraper/romrow/935/jfs_xfs_rfs_ext.html" title="fortunecity.com">http://www.fortunecity.com/skyscraper/romrow/935/<nobr>j<wbr></nobr> fs_xfs_rfs_ext.html</a fortunecity.com>

#

Exactly. Stability is much more important, IMHO. Almost nothing messes up your day faster than finding out you've lost your files because of a reboot.

#

Novelle is just CRIPPLING its product.

Just like it (and Red Hat) crippled it by removing MP3 support, etc.

#

No, they are NOT crippling their product because you can STILL SWITCH TO REISERFS if you want! Damn it, what's with you people? You bitch when they change something, then bitch again when they modify something else and then bitch some more. Get a life, fucking losers!

#

I have had to perform fsck.resierfs three times in two years all with rebuild tree in the past year and a half. This never happened with ext3 (3+ years), and it happened again this weekend, so now ext3.

Twice it occurred when a USB device did not unmount clean, and next boot was the root prompt to log in and repair the filesystem. It could not be repaired this time, well, it was repaired in that fsck stated no problems, but whatever it had to fix hosed the system. kernel panic.

TripleII

#

Never needed to use fsck.resierfs (myself,... ie not at boot) in over 6 years of using resierfs (on multiple boxes).

#

Ha! I doubt you even use ReiserFS, since you can't even spell the damn name right. Troll elsewhere, jerk.

#

The comparison between Linux 2.4.5 and Windows 2000 Professional SP1 at

<a href="http://www.fortunecity.com/skyscraper/romrow/935/memxfer5b.html" title="fortunecity.com">http://www.fortunecity.com/skyscraper/romrow/935/<nobr>m<wbr></nobr> emxfer5b.html</a fortunecity.com>

might also be interesting to some.

#

Since Hans Reiser was recently <a href="http://sfgate.com/cgi-bin/article.cgi?f=/c/a/2006/10/10/BAGERLM3RR15.DTL" title="sfgate.com">arrested in the case of his wife's disappearance</a sfgate.com>, it sounds like it may not be a bad idea.

But, why use ext3? Why don't they use Novell's NSS(Novell Storage Services)? It's fast, reliable, allows block level access for backups etc, has built in deleted file recovery, and lots more.

#

Future development prospects and a software's development path are, in practice, usually more important than immediately gratification.

I have seen actual projects that end up needed complete or partial rewrites of a lot of code several years down the line because the amateur programmers who made it originally chose to use APIs that ended up depreciated. I've had to fix some of these types of projects, one of which dependant on a bunch of Java multimedia libraries that were closed and depreciated.

They were nice at the time but ignoring long term support (something a lot of amateur programmers do) comes back to bite you in the ass later.

Gus3 made the point best: Switching from Reiser3 to 4 requiring a reformat is not an immediate disadvantage because so will switching to ext3. But Ext3 has a history of not requiring reformats to accomplish an upgrade (see ext2 > ext3) and are designing Ext4 with this same compatibility in mind. Meanwhile, the requirement that Reiser demands a reformat to upgrade to a newer version of itself makes it pretty clear that simple upgrade paths aren't as big a priority to ReiserFS as hardcore performance.

And it shows: Reiser most certainly does kick ass in benchmarks, and probably in realworld performance. But that's irrelevant, because what SuSE wants is long term support and an upgrade path, and SuSE wants that because that's what SuSE's customers want.

If that's not what you want, and merely switching the FS on install isn't enough for you, than SuSE isn't for you; there are lots of distributions that would suit you better, in fact, this is why we have distributions! People complain a lot about the negatives of having distributions, well, there's a positive if you'd like to take advantage of it.

Meanwhile, trying to use the performance as an argument against SuSE's decision is like saying that paper is inheritly better than pizza because you cannot turn over a pizza and eat it again. It's irrelevant.

#