Submitted by Andrew Morton has provided a short list of features and fixes that will be included in the 2.6.14 kernel which include OCFS2, Relayfs, v9fs, FUSE, pcmcia and scheduler updates among other changes including hundreds of bug fixes.
Andrew Morton has provided a short list of features and fixes that will be included in the 2.6.14 kernel which include OCFS2, Relayfs, v9fs, FUSE, pcmcia and scheduler updates among other changes including hundreds of bug fixes.
“reiser4
Stuck. Last time we discussed this I asked the reiser4 team to
develop and negotiate a bullet-point list of things to be addressed.
Once that’s agreed to, implement it and then we can merge it. None of
that has happened and as far as I know, all the review feedback which
was provided was lost.”
Sad. Potentially we have the best filesystem in the word but we can’t use it.
Of course i meant “in the World”
We need an editing feature, please.
I’ve got great news! You can use Reiser4 right now! You don’t need Andrew Morton’s permission:
http://www.namesys.com/install_v4.html
Get cracking!
So, take heart, you *can* use the “best filesystem in the word”.
I don’t know if reiser4 has improved lately but it was not very efficent last time I tried it (around 2.6.8 – 2.6.9). It just used too much CPU so the overall performance was degraded. And using the CPU for other thing caused the filesystem performance to drop. That was on a athlon-xp 2080mhz.
Reiser4 is more hype than anything in my experience. I was a relatively early adopter (pre-reiser4progs-1.0) and installed Reiser4-only systems (besides /boot and swap of course) on 4 or 5 machines since. I have noticed a few area of performance increases, particularly navigating trees of small files, like Gentoo’s Portage tree.
However, there are disadvantages that far outweigh the advantages. First, as the parent poster notes, the CPU utilization is unprecedented for a local filesystem. Under full load (several compilation threads, for example), the filesystem can actually starve for CPU cycles and crash the entire system. You can bet that, due to the fragile structural coherency, whatever part of the filesystem was active at the time of the crash (/var/tmp in this case), will be corrupted, and you will have to use the most severe of the three fsck.reiser4 rebuild routines to fix it. This has happened to me three times in the past six months.
Next up is the high tendency to cause to skip under light-to-moderate loads. Needless to say this became a problem for the DJ booth MP3 jukebox I set up. Reformatting as ReiserFS (v3) and reinstalling got rid of the problem.
Finally we have the myth that Reiser4 is a “low-latency” filesystem. It is no such thing. Reiser4 might be nice for a non-mission-critical LAMP-type server, but it sucks for desktop usage. It expecially sucks for laptop usage, due to the high CPU (hence: power) consumption. I’ve experienced the lowest latencies using ext3 with dir_index enabled. I’m not kidding, if you think ext3 is slow, think again. To check if you have it enabled:
tune2fs -l /dev/hdaXY
If you don’t:
tune2fs -O dir_index /dev/hdXY
e2fsck -D /dev/hdXY
For highly mobile applications where power management is important, you might want to try JFS, which has the lowest CPU usage of any of the journaling local filesystems available in the Linux kernel.
All in all I deeply regret having drank the Reiser4 Kool-Aid.
Thanks for the link
“Sad. Potentially we have the best filesystem in the word but we can’t use it.”
Reiser4 may be one of if not the fastest file system in the world however it is nothing compared to ext3’s stability and endurance, it may not be the fastest or the newest but it is in a league of its own in that respect.
XFS JFS Reiser NTFS and all the others just dont have the redundancy ext3 does and in any *real world* uses its not that much slower either – we already *have* the best imo.
are there any recent bench again xfs, jfs and reiser4?
If there were a benchmark JFS would only be there for comparison. Nobody expects it to perform like reiserfs or XFS.
I’d like to see Reiser3, Reiser4, XFS and ext3 performance compared with NTFS, UFS/ZFS and BSD’s filesystem.
It would also be nice to flex the linux device mapper by comparing compression and encryption as well.
I imagine NTFS might be able to outperform ext3 without compression or encryption, but Linux would pull ahead when its fully loaded.. or at least that’s what I might expect to see in such a comparison.
“If there were a benchmark JFS would only be there for comparison. Nobody expects it to perform like reiserfs or XFS.”
JFS isn’t too far off, and nobody expects any of the Linux filesystems to perform like ext2. JFS2 is really neat, but unfortunately it’s not likely to be open sourced anytime soon.
“I’d like to see Reiser3, Reiser4, XFS and ext3 performance compared with NTFS, UFS/ZFS and BSD’s filesystem.”
Yeah, I would, too, but any benchmark would be influenced by the variance in I/O subsystems as well as by the variance in filesystem performance.
“I imagine NTFS might be able to outperform ext3 without compression or encryption, but Linux would pull ahead when its fully loaded.. or at least that’s what I might expect to see in such a comparison.”
I wouldn’t expect NTFS on Windows XP to outperform ext3 on Linux 2.6. The Windows I/O scheduler pales in comparison to CFQ or AS.
Still no __d_path() function export?
Could some editor maybe toss in some information of what OCFS2, v9fs or relayfs are?
OCFS is shared filesystem accesible from the various nodes on an Oracle Real Application Cluster. You can get more info on this URL
http://oss.oracle.com/
The OCFS2 filesystem
http://lwn.net/Articles/137278/
A filesystem from Plan 9 space
http://lwn.net/Articles/137439/
The v9fs web site
http://v9fs.sourceforge.net/
The relayfs web site
http://relayfs.sourceforge.net/
Thanks a bunch!!!!! 🙂
Plan rules
“Sad. Potentially we have the best filesystem in the word but we can’t use it.”
agreed, very sad.
How about better laptop and wireless support. Linux needs major work in wireless. Lets face it laptops are big sellers more than desktops..
Wireless in Linux needs work?
I must say, the wireless in Linux is phenomenal. I find that it is much more reliable Windows.
What needs to be fixed is wireless drivers, but the kernel devs can’t do much about that… just only buy brands that do offer drivers for Linux (Linksys, Intel, etc.).
Now, one area where wireless is lacking in Linux is user interface. However, yet again this is not in the hands of the kernel devs. This is largely left up to desktop projects like Gnome or KDE.
Wireless in Linux is shxt at best. and it’s not just me but couple of my colleagues agree with that.
dont get me wrong, I love linux, but, in term of wireless compatibility and stability, linux totally falls waaay behind Windows (not sure about freebsd).
“Wireless in Linux is shxt at best. and it’s not just me but couple of my colleagues agree with that.”
I humbly disagree with you and your colleagues. The GNOME networking panel detects all available networks, I put in the WEP key if needed and click activate. I don’t know how it could improve. I use ipw2100 as my only net connection, and it stays up for months without a problem.
“dont get me wrong, I love linux, but, in term of wireless compatibility and stability, linux totally falls waaay behind Windows (not sure about freebsd).”
If you know anything about FreeBSD, you know that one of its primary shortcomings is wireless support. The devs are focusing really hard on that for 6.0.
Wireless in Linux needs work?
Yes, yes it does. Every wireless driver includes its own 802.11 network stack (there’s work on unifying behind a single 802.11 stack, but it’s not standard). Each driver includes its own suite of tools to configure it, instead of using the standard ifconfig command. WPA support is just now starting to pick up. There’s several different methods for detecting, configuring, and maintaining wireless devices (discover, hotplug, ifplug, divine, etc).
Linux wireless support (like everything in Linux) is all over the map.
Linux developers need to look beyond their little piece of code to see how others are handling wireless. For instance, FreeBSD has modularised ifconfig, and all the wireless drivers add their little piece to ifconfig. Meaning, a single command is used to configure all networking chipsets, whether they are wired or wireless.
More on FUSE here: http://kerneltrap.org/node/5627
Glad it finally got merged.
There was a major jump in SMP capability when Linux moved from 2.2 to 2.4. The same thing happened when it went from 2.4 to 2.6. At the moment, Linux can scale well into 16 processors. After that, there is major degradation on a true SMP system (keyword: Symmetrical). However, I haven’t heard of much work going into Linux scalability anymore. Also, with the 2.6 kernel being always a work-in-progress, I’m not sure if anything major can be done. Is there continued work in this regard?
A “true SMP system”?
You mean, as opposed to a NUMA multiprocessor system?
Hate to break it to you mate, but there are no 16-way or greater “true SMP systems” out there.
So, ignoring “true SMP” and “keyword: Symmetrical”, Linux scales very well on much larger systems than that. Where do you get your data from? What I see are companies like IBM, HP and SGI regularly testing and tuning kernel performance on 32 to 512 way systems.
I’ve been having trouble getting a Logitech Quickcam Orbit installed on SUSE Linux and heard the reason is the Kernel maintainers had some sort of fall out over the pwc module being included with the Kernel so they dropped it. Any chance this will be resolved in the near future so end users don’t have to manually install it?
The disagreement cannot be resolved as it would require implementing binary-only code into the kernel. This will never happen as the linux kernel needs to remain 100% open source. I too have a Logitech Quickcam (not Orbit, but Notebook) but I believe the driver is the same, qc-usb. I find compiling the module works better if performed manually (./configure, make all, make install) instead of using the install script.
Of course, I find compiling anything using Slackware (my preferred distro) works better than in SuSE.
“The disagreement cannot be resolved as it would require implementing binary-only code into the kernel. This will never happen as the linux kernel needs to remain 100% open source.”
I do not believe this to be true. Please keep in mind I am not a developer so will not have the terminology licked…
I was of the understanding that there was a “hook” that was not binary only but was soley there for the purpose of allowing a binary only module to be inserted, it served NO other function.
I’m not going to get in to the merits of whether or not it should have been kept in, I just wanted to point out I do not beleive the kernel needed any binary only code (for the record, the camera DID and DOES work with out the plugin, just no where near as adequately).
Yep, you’re right. It was the hook to binary code that is the problem. I forgot about the “hook-to” part of it.
But no, the kernel developers aren’t going to budge on this one. It is as it is and that’s how it’s going to stay.
i seem to recall someone else took on the pwc development and managed to remove the reliance on the binary pcwx module, but it’s still not back in the kernel tree.
i used it once (think it was v10.4 or something) and it didn’t seem to degrade in quality or functionality at all like pcw does without pcwx.
“I’ve got great news! You can use Reiser4 right now! You don’t need Andrew Morton’s permission:
http://www.namesys.com/install_v4.html
Get cracking!
So, take heart, you *can* use the “best filesystem in the word”.”
marvellous, but no good to me, i want it in the kernel by default so that my precompiled distro of choice has R4 enabled as an option by default.
>> So, take heart, you *can* use the “best filesystem in the word”.”
>
> marvellous, but no good to me, i want it in the kernel by default so that my precompiled
> distro of choice has R4 enabled as an option by default.
Most important distros do ship Reiser4 for some time now. Gentoo, Suse,…
peanut linux have had it for over a year as an option to install it / why not include it in the kernel, well it`s not 100% stable they claim.. openSUSE can include it for testing…
Just so people know there are a few reasons Reiser4 isn’t in the kernel besides bugs
1) 4K stacks: Reiser is written with the assumption that 8k stacks are the way to go. 4k saves memory and overhead so that’s the way to go, everything else works with 4k stacks now so should reiser4
2) Plugin architecture: reiser4 has a plugin architecture. Most of the kernel developers see now reason this has to be a reiser only feature.
“Just so people know there are a few reasons Reiser4 isn’t in the kernel besides bugs
…
2) Plugin architecture: reiser4 has a plugin architecture. Most of the kernel developers see now reason this has to be a reiser only feature.”
Just to clarify, the problem isn’t that the kernel devs don’t want Reiser4 to have a plugin architecture. The problem is that Reiser4’s plugin architecture falls under two categories: functionality that is already available through Linux VFS and functionality that is new and revolutionary. Reiser4 uses its own demultiplexer to handle all plugins, and the kernel devs want the low-level Reiser4 plugins to be VFS plugins instead of Reiser4 plugins. This way other Linux filesystems can use the plugins.
Hans Reiser and Namesys oppose the consensus of the kernel devs because they already have a working patch and see no further reason to dely the merge into the mainline kernel. For each kernel release that doesn’t include Reiser4, Namesys loses potential commercial client license revenues. Instead of facilitating the process of making the requested modifications to the patch (a task that will be assisted by the community), Hans and Namesys continue to drag their feet and argue with the kernel devs. The kernel devs have made it clear that they do not oppose the plugin architecture nor do they deny Namesys the right to license commercial plugins that support functionality unique to Reiser4.
There are a lot of nice filesystems ideas using FUSE:
check this: http://fuse.sourceforge.net/filesystems.html
like a relational filesytem, cvsfs, gmailFS, BitTorrentFS, SSHFS, etc
Every time a thread comes up about the linux kernel and new features therein, you can bet someone is going to start whining about lack of ricer4.
The benefits and the implementation are dubious. The kernel devs need a better reason to put it in than the 4 on the end.
“The kernel devs need a better reason to put it in than the 4 on the end.”
What about choice? And what about “many users want it”?
Are those good enough reasons?
“”The kernel devs need a better reason to put it in than the 4 on the end.”
What about choice? And what about “many users want it”?
Are those good enough reasons?”
The Linux kernel project is a republic, not a democracy. The community “selects” the tree and subsystem maintainers from the cream that rises to the top of the meritocracy. Then they decide what is best for their constituants. How does Joe User know how best to merge Reiser4 so that the filesystem and VFS subsystems remain efficient and maintainable? All he knows is he wants Reiser4. It’s like the 82% of Americans who favor a nuclear test ban treaty. Do they really understand enough about international law to make an informed decision?
If the kernel devs started merging every “working” feature that users want into the kernel, then what really separates us from the half-assed proprietary software that most of the world lives with today?
If you think you know enough to disagree with me, then you know enough to download the freakin’ patch and compile the kernel:
ftp://ftp.namesys.com/pub/reiser4-for-2.6/2.6.13-mm1/reiser4-only….
“If the kernel devs started merging every “working” feature that users want into the kernel, then what really separates us from the half-assed proprietary software that most of the world lives with today?”
And since when does proprietary software give the users what they want?
“It’s like the 82% of Americans who favor a nuclear test ban treaty. Do they really understand enough about international law to make an informed decision?”
Actually it has been proven time and again that ordinary people can be wiser than moronic governments, and if you want facts I’ll send you a private message.
How many years we must wait for inclusion Reiser4 in main kernel tree ? What is do terrible may happened with kernel if we put two extra option ( CONFIG_REISER4_FS and CONFIG_REISER4_CHECK ) ? Politics …
Very well said!
How many years we must wait for inclusion Reiser4 in main kernel tree ? What is do terrible may happened with kernel if we put two extra option ( CONFIG_REISER4_FS and CONFIG_REISER4_CHECK ) ? Politics …
No, tech. Linux is not a life support system for Reiser fs. The kernel as a whole has to be balenced as it is designed to work on everything from small embedded systems up to mainframes.
Hans Reiser is understandably frustrated, though he’s also been given a path to get Reiser fs included in the mainline kernel. He’s a smart man and has smart people working with and for him. If he can’t resolve this with the other kernel developers, he’s stuck with patching the mainline kernel.
Something tells me that there is a technical solution. What it is is not obvious partially because Hans is rightly concerned about future customers (the ones who want custom file system plugins).
Till then, I’m sticking with EXT3 personally, though Reiser fs is interesting enough that I’m seriously considering pushing it for use in a very high profile project. (I’d love to brag, though we don’t have the contract (yet?), and there’s only so much I could say anyway. Sigh!) One problem with Reiser fs is that it is currently practical under Linux, and I do not want to push Linux as a solution; process before product is critical at the initial stages, and the customer should have a way to migrate to other tech as necessary. Reiser fs if used fully restricts how you can move on to other operating systems that do not have all the nice features (metadata handling) that Reiser fs has.
Although I was a little skeptical at first, I get the sense that git is really fitting nicely into the kernel dev process. It’s really easy to check out git snapshots, maintain your own git tree, and create git patches. It seems that it makes Linus and Andrews life (a little) easier as well.
Check out the documentation for Cogito here:
http://www.kernel.org/pub/software/scm/cogito/README
What does reiser4 have what ext3 doesn’t?If i wanted i could patch my 2.6.13 gentoo kernel just as i did to get the dazuko module loaded for on-access file scanning with clamav.I think reiser is a hype and.
i agree with butters that the vast majority just wants something because it seems to be cool having it,reiser4 is a hype for a lot of people.
Yes i know,works like a charm:
localhost netpython # uname -a
Linux localhost 2.6.13-g4706df3d #1 Wed Sep 7 12:59:13 CEST 2005 x86_64 AMD Athlon(tm) 64 Processor 3000+ AuthenticAMD GNU/Linux
Actually it has been proven time and again that ordinary people can be wiser than moronic governments, and if you want facts I’ll send you a private message.
I agree,how can anybody be more stupid and or evil than a goverment who spends rather 300 billion on a lost (oil) war from day one than 20 million on the cassualties of Katrina.Would have been nice if at least half of those 300 billion would have been spend on education,healthcare,prosperity for all.
>> marvellous, but no good to me, i want it in the kernel >> by default so that my precompiled distro of choice has >> R4 enabled as an option by default.
> Most important distros do ship Reiser4 for some time now. Gentoo, Suse,…
no they don’t, SUSE at least only supports Reiserfs, or R3 if you will.