Slashdot is powered by your submissions, so send in your scoop

 



Forgot your password?
typodupeerror
×
IBM

JFS May Make It Into 2.4 191

Grimsaado writes: "LinuxWorld has an article on IBM's JFS and how it might be included in 2.4 as well as some technical fluff on it's phenominal cosmic power." Heck, with the number of journaling file systems, it's like being at a file system buffet at this point.
This discussion has been archived. No new comments can be posted.

JFS May Make It Into 2.4

Comments Filter:
  • Of course, this is only one application of Java.

    And it's the only one worth a damn .. well, sometimes worth it. Face it, Java had high hopes of being the 'platform independent' language.. now it just sucks and is slow. Caused in part by M$ muddying the waters its own different version of Java no doubt. And with Java applets and Javascript it's just a horrid horrid mess.

    PHP / ASP is what kicks ass on webpages for me. Clean, fast, efficient (well, PHP anyway - but I have seen ASP do nice things too).

    Still, that's just my opinion .. not to say that you're wrong, just that I don't agree. :)

    --
  • by Anonymous Coward
    Wasn't Linux 7.0 released just last week?
  • by Anonymous Coward on Thursday October 05, 2000 @03:55AM (#729853)
    The ReiserFS really rocks, but journalling is only secondary. The ReiserFS is much more than a journalling filesystem. It is supposed to ultimately be some sort of object oriented user-configurable file system with personality plug-ins and all sorts of bells and whistles. Me, I'm only interested in the journalling aspects at present. It is rock solid now and I won't go back to ext2 anytime soon. The only thing I miss with ReiserFS is extended attributes like immutable and append-only. These are planned for a future release of ReiserFS, however.

    If Linus has a problem with ReiserFS, it is that he probably fears some of the exciting new ``disruptive'' concepts that Hans Reiser has planned. ReiserFS is truly innovative. For those interested in the innovations, there is a White Paper [devlinux.com] available. I'm sure that Hans Reiser's roadmap is what is scaring Linus. I'll be happy to see JFS make it into 2.4, but ResierFS deserves to be there too. I urge anyone with a slight interest to try out ReiserFS. I'm sure you'll agree then that it deserves a place at the table. I'm a late adaptor, and skeptical of new code (I used Xia FS for years after ext2 was available). If conservative old me can handle ReiserFS, anyone can.

  • I was using JFS under AIX for several years. I can't imagine why people would want to run it. It doesn't give you much more data security (only file system structure is journalled), and you pay a heavy price in terms of performance.

    Does the JFS on AIX do synchronous or asynchronous writes? I know they say that the linux version is async.

    Fsck on ext2 is pretty fast, crashes are very rare for server systems, and servers require regular backups anyway.

    If you had, say, a 100gb+ RAID full of data, a full fsck would take a very long time, especially if it was a particularly active partition. With a JFS, as you should know, it takes a minute (maybe less) on a nice SCSI or FC RAID. Hell, Win2k's NTFS5 is (partially) journaling, and I haven't lost any data on Win2k due to a crash besides data I haven't saved. If JFS can't accomplish the same feat, then IBM isn't really trying.

    We gain more than just filesystem integrity through crashes with the full JFS package, though, because we're getting the LVM. The only question is, will we be able to have partitions grow themselves automatically, like you can on AIX? Now that would be cool.

  • GFS is the journaling -cluster- filesystem, now in Beta, that also works on a plain old non-cluster box. See the Sistina [sistina.com] site.

    -dB

  • Yep, that figures. Steven Tweedie is the author of ext3, and he views Reiserfs as a direct threat. Note that Alan Cox and Steven Tweedie are close friends. They are both brits, and both Red Hat employees. Alan Cox is running some interference for Tweedie hoping to throw a wrench into the works against Hans Reiser. He probably wants to cause problems for Reiserfs until ext3 is ready. It is a delay tactic on the part of Alan Cox, and it is exhibits a particularly nasty side of Cox's personality.
  • None of which can read/write BFS at this point, despite the fact that it's well documented in Practical File System Design with the Be File System by Dominic Giampaolo.

    Do you have a need for this? If so, and you have coding skills, you should start work on BFS for Linux.

    That's why most of these projects get started - somebody had an itch they needed scratched.

  • > the changes that were being made
    were radical enough that Linus refused to allow
    them in relatively late in the development cycle

    That's not exactly that. The changes were needed, but linus wanted an infrastructure that would support *all* the journaling file systems, and not only reiserfs. (The reiserfs patch was rather ugly too, but cleaning it wasn't sufficient)

    But such an infrastructure have been defined by Alexander Viro, mostly for ext3, that was on the drawing board at this time, and was not really perfect for reiserfs (and in fact turned out to be quite bugged too). Hans Reiser had the feeling that all there was an ongoing conspiration to push its filesystem out of the kernel until ext3 was ready, and took grief of that.

    The truth is unkown, people involved in that have bigger ego than signal 11 (if that's possible) and reiserfs 'commercial' background is probably costing it bit of support too.

    Cheers,

    --fred
  • One irritating thing about it is that theres no support for floppy disks, or other small removable media.

    I can see how you might want to use a filesystem on, say, a 1.8mb flash card or something, but I can see no reason why you'd want to use a journaling filesystem on a floppy disk. In fact, when I use a floppy to transfer files between UNIX systems (a very rare occurrence indeed) I don't even bother to use a filesystem; I just tar to the floppy device itself. Otherwise you have to go through the whole mount/umount procedure, possibly with a mkfs or an mformat in there someplace.

    Anything smaller than, say, 40mb doesn't need a filesystem on it if all you're doing is shipping data around. There are still other filesystems (as others have noted.)

    And finally, I can't see why you'd need a JFS for flash memory. You shouldn't be writing to it all that often. The odds of losing power during a write are exceptionally slim. And SRAM is fast enough to where it shouldn't be a big problem, but I can almost see a need for a JFS on SRAM. However, in an application where you don't have a hard disk, you probably don't have the RAM to spend on the LVM anyway, so you'll want to look more at something like ReiserFS or ext3 than you will want to examine something as large as JFS, anyway.

  • by Anonymous Coward
    Point me at one useful succesful web site that makes any use of all this third party crap.

    Admit it - Sun and all of their third party partners are just hoping if they glom together enough crap, the ensuing mess will somehow appeal to the less-enlightened by virtue of sheer hype.

    As for "addressing scalability" - I can tell you from personal experience building sites that get 100 million+ hits a day, that Java isn't even on the option list for truly high volume websites.

  • Dear fauxpas,

    No no, it's too obvious.

  • Do you have a need for this? If so, and you have coding skills

    Yes. Not really. If I can get help from somebody competent, then I'd do it.

  • Dude!!! If your configuration doesn't work and everyone else's seemingly identical configs do, then you screwed up. Claiming otherwise is FUD.

    I think you both need to build a bridge and get over it. For all either of you know, ReiserFS and dosfs (or whatever it's called) had an interaction with a third piece of software (a driver, possibly) which took exception to some string in some PROM for a hard disk or similar.

    Kernels are frightening and twisty places to be, linux's kernel is no exception, especially since there's so much code from semi-random sources. I'm not trying to malign linux or linus or reiser or your mom or anyone else, but I'm just saying that there's room for both of you to be essentially right here.

    Oh, and BTW, xconfig should either be removed from the kernel distrib or made to work religiously. IMO menuconfig and xconfig should just be symlinks to the same TCL script, and when you call it with xconfig it should use Tk, but I guess that's just my opinion. If that were the case, then there wouldn't BE this problem of one config tool working, and one not.


  • Dude!! It didn't work on my system when the *only* change was the inclusion of reiserfs, giveing a very good indication that the problem was with reiserfs... Claiming otherwise is FUD.

    Just because it works for your roommate with one setup does *not* mean it will work for every other setup. It's quite possible he's running an older version, a newer version, or a different test kernel.

    It's quite obvious that reiserfs is not ready for inclusion the kernel code.

    Ranessin
  • Well, IBM's "JFS for Linux" is based on their port of JFS for OS/2 (remember that?) as opposed to their JFS for AIX, which is what most people associate with the name.

    The current JFS for Linux project is, for instance, still case insensitive. Hardly an acceptable situation for a UNIX filesystem, but hopefully one that can be fixed.
  • "I think you both need to build a bridge and get over it. For all either of you know, ReiserFS and dosfs (or whatever it's called) had an interaction with a third piece of software (a driver, possibly) which took exception to some string in some PROM for a hard disk or similar. "

    Thank you. This just goes to show that till these quirks are worked out, reiserfs is not ready for the primetime.

    Ranessin
  • And guess what ? If it's obvious, odds are I've heard it before.

    At least you seem to know it's not pronounced Fox Pass. You're a rarity in that respect.
  • >What's really sad is that it's the many casual linux users -- people
    who would use reiserfs if it were in the kernel, but are afraid of
    kernel patches, or don't have the time to figure >them out -- who are paying the price.

    It is not the fault of the kernel developers that Joe (L)user does not know how or have the time to patch the kernel.
    This is a good chance to learn about patching the kernel.

    Mojo
  • AFAIK JFS is both a journaling and a logging file system. Dunno what's the difference, but I remembered of this point, because they (IBM) said, this would be something special.
  • by onion2k ( 203094 ) on Thursday October 05, 2000 @03:57AM (#729870) Homepage
    Theres more about this thing at http://www-4.ibm.com/so ftw are/developer/library/jfs.html [ibm.com]. One irritating thing about it is that theres no support for floppy disks, or other small removable media. Understandable considering the system, but floppy disks are still quite handy. (Assuming the page I was reading is up-to-date)
  • by Greyfox ( 87712 ) on Thursday October 05, 2000 @03:57AM (#729871) Homepage Journal
    Disclaimer: Since the article is slashdotted, I haven't had a chance to read it yet. That being said...

    Sure, let's just piss in Hans Reiser's petunias. A lot of people I've talked to seem to thing that Reiserfs is the farthest along journaling filesystem and I'm sure including some other journaling filesystem in 2.4 would be a major poke in the eye for him.

  • by shippo ( 166521 ) on Thursday October 05, 2000 @03:58AM (#729872)
    Adding any journalled filesystem to the existing kernel requires significant changes at the VFS layer. Making such a change has the potential of breaking other things, particularly other file-systems. Every current filesystem will have to be rechecked - not a simple task as some don't have full-time maintainers.

    The current 2.4.0-test kernel is getting very close, and Linus now appears to be only accepting bug-fixes and the odd self-contained driver. There is no way such fundamental changes could go in now.

  • I cannot read the article (as usual - slashdot effect), but from what I read from the IBM web pages, the JFS will be released under "an OSI approved license" - which means in plain english - not GPL, which means - Linus will not include it. period.

    I also read that this JFS is a sort of a "cut" version from the full JFS that AIX have.. Anyone from IBM can shed some light on this issue?
  • by 11223 ( 201561 ) on Thursday October 05, 2000 @04:01AM (#729874)
    Heck, with the number of journaling file systems, it's like being at a file system buffet at this point.

    None of which can read/write BFS at this point, despite the fact that it's well documented in Practical File System Design with the Be File System [barnesandnoble.com] by Dominic Giampaolo.

    Grumble grumble complain complain...

  • Seriously, haven't you ever sat in on a job interview with a geek? When Java is mentioned, he suddenly bursts into, "Well, I can make your webpage shine and sparkle, I can make the logos spin, I can have interactive pull-down menus, I can have the logo turn into a smiley face whenever the user hovers his mouse pointer over the banner ad..."

    Well, what can you do about setting up a database for us with your Java experience?

    [Dead silence]

    The biggest problem with Java is not with Java itself; it's with the naïve software engineers who only want to do "neato-cool" things with it. This makes Java engineers look more like effeminate interior designers.

    Another problem with Java is its inherent latency. Java servlets perform, on average,at one fourth the speed of a comparable perl or ASP database. For instance, Winamp's entire server is run on Java servlets. It currently runs half as fast as Slashdot (which we already know, is kinda slow already). Add AOL's outdated unix server, a slow client computer, and an even slower connection, and you've got a really bad experience on the client side.

  • Java Servlets for instance is very widely used as the backend for a lot of websites...

    whose webmasters don't understand that there are vastly superior ways of building websites.

    Really, Java has its uses, but JSP has got to be the most fractured and counterintuitive technology I have ever come across. Even on the most ra-ra Java websites, you can easily find serious critics of JSP.

    Compare JSP to PHP generated pages for any application - there's no comparison with regards to simplicity, speed, and suitability.

    Hell, even mod_perl is preferrable.

  • Calm down, man. You're getting all bent out of shape over a flame that wasn't even there.

    Let's review: The original poster mentioned a rumour that was going around in hopes of getting clarification. He got that clarification along with some slightly-outdated information. And then you went apeshit over some perceived attack on IBM.

    A simple correction would have sufficed.
  • If you had, say, a 100gb+ RAID full of data, a full fsck would take a very long time, especially if it was a particularly active partition.

    Well, first of all, people routinely use ext2 with 45G or 60G drives (I have one), and it doesn't seem to be a problem. Furthermore, boot time can't be such a big issue for the kinds of customers IBM is targetting because many of IBM's AIX machines used to take from minutes to hours (!) on every boot just to get their SCSI subsystems up (I hope they have improve this by now); in comparison, the time for any fsck is negligible.

    But most importantly, just because JFS spreads out the time for maintaining file system accesses doesn't mean you aren't paying for the time. Each day, the transactioning that JFS may cost you a few hours in computer time, compared to a system that doesn't do transactioning. And with JFS you pay that cost whether your system crashes or not. In fact, the more reliable your hardware is, the higher the cost of JFS.

    So, would you rather lose a few hours of computer time every day, or lose a few hours of computer time on the very rare occasion that the machine was not shut down properly?

    The only question is, will we be able to have partitions grow themselves automatically, like you can on AIX? Now that would be cool.

    "Cool" maybe, but not very useful. Or are you in the habit of leaving most of your disk unpartitioned so that you can eventually grow into it? Last I used it, LVM/JFS couldn't even shrink file systems.

    I think the popularity of JFS/LVM on AIX is rooted in particular idiosyncracies of IBM culture and limitations of the AIX operating system. On Linux, easy backup/restore, GNU parted, fast fsck, and fast boots give you more flexibility than LVM and JFS, with less runtime overhead and less complexity.

  • actually it started out as a correction. I went apeshit later while I was typing it and let it kind of get out of hand.
  • So. I can't install linux on my home machine. And yes, I've tried various distributions with various (2.2+) kernels. So I have screwed up. Or could it be that something is wrong with my hardware. Never mind that my machine runs Win2K beautifully? It seems to be a memory problem, but various tests indicated (MemTest 86) that there is nothing wrong with the memory. It also could be the maxtor harddrive, I may never know. So no Linux on my home machine. And what the heck, I use my home machine only for games.

    And having tried to install Redhat 6.2 (and 7.0) on a work machine. Neither of the installs did succeed. No one I knew could figure what went wrong. But the routing (kernels IP routing) just went trought the roof.

    Yet I succeeded in installing SuSE 6.4, and it worked. I finally changed SuSE to Debian and it's apt-get delicacies.

    Think about FUD, because not everything that goes against some things you hear or believe is FUD. Some things just break mysteriously on various places.

    OB-Topic: I just can't wait for decent journalling filesystem in kernel. The LVM is already being shipped in Debian...
    --
    when everyone gives everything,

  • by ssimpson ( 133662 ) <slashdot@samsimps o n . c om> on Thursday October 05, 2000 @05:31AM (#729881) Homepage

    Scramdisk [clara.net] is currently being ported from Windows 95/98/ME & NT/W2k to Linux (by myself and AJ). This will allow the creation of a "virtual container" that can contain any filesystem - including filesystems that implement journaling.

  • One filesystem doesn't.
    you fudged up.
  • <blockquote>Bottom line: ReiserFS gets two thumbs up. Highly recommended if you're not afraid of patching your kernel.</blockquote>

    Alternatively, you could just use SuSE, which lets you select ReiserFS when you install, no patching necessary.

    As an aside, the USB works like a charm, too. I'm using my Handspring Visor and a Wacom Graphire without trouble. Well, except that I can't seem to get the active area and my screen resolution to match...

    I'm doing all of this on my laptop and the ReiserFS is just great. There is even a noticeable difference in speed when entering a large directory in mc.

    Now I'll have to get SuSE 7.0 and install it on my desktop machine. Fsck'ing an 8GB partition (the largest one in there) is definitely not fun.

    Really, I want to have extended attributes like in BeFS. ReiserFS promises that but it's not there yet.
  • Isn't the Reiser File System the most advanced of the journaling file systems? Linux already stated that it would NOT make it into the 2.4 kernel.
  • > This isn't a horse race - the guy with the first journalling file system doesn't get a check for $100,000 USD

    Unfortunately yes. There is strong incentive behind the journaling filesystem, as it is a business requisite. Hans Reiser have a commercial buisiness to run, and if rfs is the one true linux journaling system, it would help its aggeda a lot.

    Don't think that there will be space for 4 journaling systems. Think about ext2. Everybody switched. Everybody will switch to a journalling file system. Today, rfs exists and work, and if it got into the kernel say 4 or 5 month ago, the journaling war would be over. (Would you reformat all your disks to try another file system ? It may be done to get journaling, but not for the pleasure).

    Cheers,

    --fred
  • Dearest lad, I was raised in the finest schools and ate off tibetan monk bellies cleansed in spring water. I know my pronunciation however rusty the spelling may be.

    "F-oh pah"

    Anyway, I like the cut of your gib, and knight you Sir Faux Pas.

    Rise Sir Faux Pas.

  • Don't denounce me for not knowing; I plead ignorance since the page can't load. But I hope JFS doesn't stand for what I think it does. (Java File System).
  • The JFS in AIX is all in assembly and is completely unportable, thats why the OS/2 version is being used.
  • Before you spout too much about kernel configuration, I'd suggest grabbing ESR's CML2 configuration system [tuxedo.org]. This has all you want and more, the only possible problem being its reliance on Python (last I checked). It's very well thought-out and deserves acceptence into the kernel sometime soon.

    I have used Reiser in the past, but some big fs corruption due in part to a badly-applied Reiser patch (I think...) forced me to reinstall, and I haven't gotten Reiser back in yet. Unfortunately my partition table is full at the moment (everything primary -- dang x86 partition table), so I can't even resize something to make room for a temporary backup partition. Anybody know how to make a primary partition become a logical partition, especially if resize is a possibility?
  • It is all-you-can-eat.

    BUT this is good! This is called progress, and this is called competition. File systems may seem moot, but I'm glad people haven't given up. Who better to poke the fire than IBM?

    I think the future holds many things in file storage (perhaps we'll be using XML for data structure), and the more I hear about the basics of storage, the better I think computers will become.

    ----

  • Nope. Journaling File System. Another in a series of a series of a series of a series of a (thwack) of a series of journaling file systems due to make it into the linux kernel.

    Others include (from memory)
    ext3
    xfs from SGI
    ReiserFS
    I think there are others, but I can't remember.
  • No, it looks to me like XFS is much more advanced. At least, it has significantly higher limits (filesystem size, file size, number of files, etc) and ALL of the required features of an ideal JFS.
  • The page is still slashdotted. In the JFS for Linux-FAQ's [moelabs.com] it says "We were able to get JFS on 2.5 list of items to be merged....", however, the JFS core team wants to wait until an alpha version is ready (a beta version is excpected ro be released this month) before any real attempts to make the JFS part of the (standard) kernel are made.
  • > it doesn't matter if you are busy, employed, don't know C or C++, or
    > that the person telling you this has never even seen the kernel source code.

    Well, I am busy, I am employed, I do know C and C++, and as a matter of fact, I'm reading up on kernel internals and working on a little hax0r that I think will be a very cool addition for the next developmental round if I can get it working. Now, if only Slashdot supported a killfile so I could say *plonk*
  • Wasn't Linux 7.0 released just last week?

    That was Ret Hat's distro number, which is different from Linux itself.

    ...like I said, though, there's some room for confusion...

  • I had seen a demo of XFS and JFS about 6 months ago. At this time I was already using ReiserFS. At that point XFS and JFS in my opinion had a very long way to go especially XFS. ReiserFS has seemed to be stable for quite some time now I have been used it on production machines without a problem. It has also been included in some distros like Mandrake 7.0. This seems to be a proven product in my eyes. I would hate to see a good product passed up because there is political or commercial (insert IBM logo) motivation. Also not to leave out ext3, I have not used this but here that it is also very mature at this point. I know some commercial companies are using it for their SANS such as VA. They have been using a modified version of ext3 for thier SANS for at least 6 months now. Also seems proven.
  • Tough I'm not a specialist at all, I read an interesting article [linuxgazette.com] on it at Linuxgazette [linuxgazette.com]. Interesting technical information (datastructures etc...)
    By the way, I got the link in a comment here at slashdot (some time ago).
  • You don't happen to be any relation to that Mahir fellow now do you?

    He robbed me of three shipments of gold, the giddy-legged swine.

  • NTFS is not a journaling file system. Where do you get your information? Probably just wish it up. NT mounts a dirty file system and boots, while in the background it runs the file system fix up code. Truely a bad idea.

  • You are mistaken. JFS is fully journaled, both data and meta-data are journaled. Reiserfs only journals meta-data.

    JFS is a good choice for business systems where the data is extremely critical, like financial data.

    Troy
  • by SEE ( 7681 ) on Thursday October 05, 2000 @10:35AM (#729901) Homepage
    Actually, he heard right ;-).

    You see, while JFS is available for OS/2 Warp Server for e-Buisness (Aurora, Warp 5 Server), it's not available for the workstation version of OS/2, OS/2 Warp 4.5 (Merlin with the kernel update fixpack).

    So a project was put together to take the GPLed JFS code and get it to run on OS/2 4.5.

    Steven E. Ehrbar

  • > Not allowed to.

    Huh? Learn how to join the kernel mailing list

    Been there. Done that. All of the decisions are actually made off-list. The list is a decoy.


    > And then ignore your patches.

    Hmmm, maybe that says something about the quality of your coding... (i.e. it sucks ass)



    And who are you, exactly? Some expert in ass-sucking? Getting patches and ideas rejected by The Posse (Viro, Cox, Tso, Molnar, etc) doesn't mean they're bad. To date, it's meant that they tickle an ideological allergy to things not posix. For instance, even though Linus stated he wanted a clean way to suppor streams in Linux, in order to support existing filesystems, The Posse wouldn't let it happen, and even went out of their way to stifle the debate. Ask Cox about his kill file sometime. He won't even include fixes to printk to provide 64-bit support (in spite of the fact that linux is supposed to run on 64-bit machines, like UltraSparcs).

    Just because they're the current in-crowd dosn't make them right. For instance, Linus refuses to let a kernel debugger be included in Linux. He, and a number of morons on the kernel list, say that printk is all you need to debug the kernel. Yeah, okay. But then support for printing 64 bit numbers in printk is rejected, meaning that it's actually impossible to debug 64-but data structures with the recommended method, printk. It's all sort of silly. I can't wait to see 2.4.0-pooch-screw-37.

    I can understand that Linux doesn't have a design, that's it's evolved as it's coded. But it could at least have a philosophy. Currently (2.4.x) it just has problems, and the Mindcraft benchmarks to refer to.

    ________________________________________
  • I'm LOSING karma with this one, not gaining it. I do need some deficit, otherwise I'll make an Icarian flight.
  • by shaggykl ( 187717 ) on Thursday October 05, 2000 @05:42AM (#729908)
    I'm from IBM.

    As previously stated JFS is released under GPL.

    AIX's JFS contains licensed code from outside sources. Several years ago, JFS was redesigned from the ground up for OS/2 Warp Server. This version does not contain any encumbered code and was designed to be more scalable than AIX's version. This JFS first shipped last year with Warp Server for e-Business.

    Therefore, the Linux offering of JFS is not the same filesystem you'll find on AIX, and you won't be able to share a JFS file system between AIX and Linux. (You will be able to share one between Linux and Warp Server.)

  • The 1/4 figure comes from the servlets at www.winamp.com taking 10 seconds to load, as opposed to the Slashdot perl scripts taking 2.5 seconds to execute.
  • by Bazman ( 4849 ) on Thursday October 05, 2000 @05:55AM (#729914) Journal
    Thats my biggest fs problem at the moment. The average home user can go make a cup of coffee while the machine fscks after a crash. Even our departmental server doesnt take that long. But just about the only time I've ever had to restore stuff from backups is when people have done

    rm * .tex

    or similar. I do all this backup nonsense just to protect themselves against their own stupidity?

    So, are any of the journalling FSs smart enough to rollback a journalled transaction to undelete a file? And provide user-level tools to do it?

    I know there is the beginnings of undelete support in ext2 FSs, but its all very beta. Surely when designing a new FS you'd factor it in from the start...

    Baz
  • Yeah, they'll add it by "accident" like with JFFS.

    Hi, we're happy to announce the release of kernel 2.4.0-pooch-screw, wherein we screwed up the VM and VFS again, and occasionally even Ext2 gets scragged, but we tossed in another journaling filesystem.

    Perhaps these guys are smoking cherry-flavored crack. What's the kernel list have to say about this? Viro? Care to chime in? so they're working with "the community" to get it included, in spite of the fact that the thrid feature freeze is on?

    ________________________________________
  • by TheTomcat ( 53158 ) on Thursday October 05, 2000 @04:16AM (#729916) Homepage
    Heck, with the number of journaling file systems, it's like being at a file system buffet at this point.

    Mmmmm.. all you can eat inodes.
  • If /. did mirror sites, then that could take ad revenue away from the said site. There is also copyright issues.

  • This causes filesystem checking software (like fsck) to work dramatically faster.

    fsck does not go faster; it is simply not needed. The file system guarantees that it will always have its integrity intact. Therefore there is no cleaning up to do after an unclean shutdown.

    There are still fsck programs for journaled file systems, though - they might be useful for example after being hit by a bug in the file system itself.
    --

  • by Anonymous Coward
    From a practical standpoint, Minix and DOS file systems are the best for floppy disks. Name length can be a problem, but I believe that some versions of Minix can handle 32 character names. It really doesn't matter if you tar everything up before you put it to floppy. You probably can tar directly to the floppy without any file system: tar cvf /dev/fd0 foo bar (assumes formatted floppy!). More advanced file systems require so much overhead for journals and superblocks that there isn't much room left for data.
  • by dpilot ( 134227 ) on Thursday October 05, 2000 @05:57AM (#729925) Homepage Journal
    Bingo!

    This came up months ago on the kernel list - simply the weekly summary, even.

    Journalling requires changes in the VFS. Rather patch the VFS many times for each member of the 'buffet' of journalling filesystems becoming available, Linus said he'd prefer to find the common elements, and make VFS "journalling-ready".
    The individual journalling filesystems would have to work with the new VFS to make sure it was suitably changed, and to make sure their code would work with it.

    This sounds like the correct approach to me, even if it does delay things a bet. Better than letting ad-hoc adaptations creep into the kernel.
  • JFS is also available in an OS/2 based workstation client from Serenity Systems called eComStation. See http://www.ecomstation.com or http://ecomstation.mensys.nl

    This workstation software not only includes JFS, IBM Desktop on Call (remote control software), Star Office and Lotus Smart Suite .. but the same SMP which is in Warp Server for e-business is also available as an option.

    Not too shabby.

    Regards,
    Bob St.John

  • Does eComStation (eCommStation?) have a market beyond OS/2 upgrades?

    All I hear about it is a (sort of) upgrade from OS/2 Warp 4. But it could be because I am not in the circuit of huge parks of company computers.
    __
  • So I'm not the only one getting that impression. It seems to me, too, that he's getting the runaround. "Oh, I'm sorry Hans, but we can't possibly put your filesystem into the kernel until we get this fancy new VM layer finished." Never mind that reiserfs has been working beautifully in the 2.2 kernel for a long time, and probably works as well in 2.4 (I haven't tried it).

    I've been using ReiserFS since 2.4.0-test4 or so and it's been working just fine, with test9 it's just amazing.

    My impression is Alan is keeping it out of the kernel because they don't want support costs going to the ReiserFS company, but want that all to go to Red Hat for ext3, nevermind that ext3 is such an immature, unusable peice of.. well, nothing really.

    I was pretty upset about ReiserFS not getting into the kernel before 2.4.0, Linux absolutely NEEDS a Journaling Filesystem, and ReiserFS works beautifully and is the most mature.

    -- iCEBaLM
  • Hmm .. is Scramdisk Linux opensource then? I use e4m [e4m.net] since the source is available. I never use any crypto products where the source hasn't been submitted to open peer review ..
  • By 2.4, I think that they mean 2.4.x and not 2.4.0

    I agree that they won't get JFS into 2.4.0, especially since there's a feature freeze and even a bug-fix freeze except for fixes to serious bugs. The article also mentions that ReiserFS is likely to be finished & into the kernel before JFS. I believe that Alan Cox proposed an abstract journaling layer in the kernel that the various JFS's can plug into. This is probably going to be necessary at some stage as there now seem to be at least 5 JFS's for Linux: Reiser, IBM JFS, SGI XFS, Ext3 and TUX2.

    HH

  • by Azog ( 20907 ) on Thursday October 05, 2000 @04:26AM (#729939) Homepage
    I agree - it won't be going in. I read the Linux Kernel mailing list, and I haven't seen a single post on JFS in the last week.

    It seems the biggest problem at the moment is last minute changes in the VM - Ric Van Riel has rewritten parts of the VM to be much faster, but there are some deadlock problems and other bugs being worked out.

    On the other hand, many people are testing prerelease versions of 2.4.0 with the ReiserFS patches and not having problems. Even if ReiserFS doesn't make it into the official kernel release it will probably continue to be a "standard" patch and available in many distributions, such as SuSE and Mandrake.

    There are already big improvements slated for the 2.5 series - a cleanup of all the IDE code, for example.

    Torrey Hoffman (Azog)
  • by Noodles ( 39504 ) on Thursday October 05, 2000 @04:28AM (#729940)
    Thanks, Hans.
  • Linus stated that it will not be included in the 2.4.0 kernel. And even according to the article (beginning, paragraph 4):

    I know, the ReiserFS is much further along, and will almost certainly be the first journaling filesystem included in the 2.4 kernel.
    Revelation Zero: The beginning of the end.
  • "It takes much of the power of C++, adds an excellent standard library, and lets you ignore the low-level and worry about design. And it's the most purely OO language in use."

    As any yacht captain will tell you, you'll never lead the pack with a full load of bilgewater. The same is true in programming languages: if you don't get down-and-dirty with the low-level, the end result is guaranteed to be inefficient. This is why the naïve Java programmers can never build anything low-lag with a JDK.

    "If Java is so "inefficient", I'm curious why it's supplanted Perl and ASP in both the UNIX and NT environments I've worked in."

    Okay, reality check time: Java is at the bottom of the database barrel as far as performance and reliability go. Just see winamp.com's database structure (that is, if the site is running at all!). They use Java servlets. The average latency for a query on a 60K/sec connection is 10250ms; four times as slow as PHP, ASP, and Perl.

    My suggestion to you is to take those blinders off, stop trying to say "Sun Solaris 7!" with the effeminate lisp, and start learning some worthwhile database APIs.

  • It will be marketed beyond the current OS/2 user base. But the initial market, the "early adopters" are certainly expected to be OS/2 users.

    eComStation itself is part of a larger set of managed clients, to include support for Windows and Linux, which are likely to be available in first quarter 2001.

    However, even the managed client with the OS/2 engine will be generally marketed, though it's probably more likely to gain acceptance outside the US before it gains significant popularity inside the US.

    Regards,
    Bob St.John

  • by NetJunkie ( 56134 ) <`jason.nash' `at' `gmail.com'> on Thursday October 05, 2000 @06:16AM (#729956)
    Someone has changed RedHat 7.0 to allow you to install ReiserFS during boot.

    http://cambuca.ldhs.cetuc.puc-rio.br
  • Arguing with you is amusing. You change the topic of the discussion and then flame for something that have not been said.

    That criticism was... random.

    Considering the topic of the discussion is Hans Reiser and the inclusion of his journalling file system, I think making the point that, according to those who actually work on the kernel, the kernel isn't ready for a journalling file system somehow does have relevance.

    The point is that there *is* money behind journaling on linux, and that it *is* a horse race. And btw, the kernel is ready for it, as it already exists as a patch. The kernel is not ready for the inclusing of a journalling layer, which is a very different matter.

    No, it's the same matter. You're simply trying to look at a distant object and focus on the tip of your nose so that it appears as two distinct issues (if you don't know what I'm talking about, try it).

    The journalling layer is what makes the kernel ready for journalling, as far as those who work on the kernel go. Therefore, in their opinions, until that journalling layer is ready the kernel is not ready for journalling file systems. That ReiserFS exists now, and works with the current kernel setup, in no way alters the fact that the kernel maintainers don't consider the kernel ready for journalling file systems. That, in the end, is the only point that matters from a technical standpoint. And that's why ReiserFS is not going to end up in the standard kernel source.

    At any rate, the existence of a working (and from what I hear, very reliable and stable) ReiserFS indicates that Hans Reiser has already won the journalling file system "race." Business can already use the fruits of his labor. He can already proclaim that his product, and his energies, resulted in the first stable, production-quality journalling file system. He won, over IBM, SGI, Stephen Tweedie, and probably several others. Congratulations to him. Hans Reiser's agenda of being the first to a stable journalled FS has already been fulfilled, and he can get the money, and the accolades, and whatever else is necessary to satisfy his ego. He can already demonstrate it, use it, sell it (or whatever he wants).

    This crap, however, about being the "standard" file system is just that: crap. Does he really think journalling file systems are going to be the default file system in the new kernel? (The default is going to be ext2 for a long, long time to come.) Being included in the mainline kernel is nothing more than a timesaver from having to download his patch, and anybody who understands the merits of a journalling file system isn't going to be dissuaded by downloading and patching a kernel (does he really think a competent sysadmin is going to be frightened of obtaining a kernel patch?). As a result, getting his code into the kernel is pretty darn close to being a meaningless accomplishment.

    That's why I don't understand why he sees getting into the kernel as such a huge accomplishment, especially at the expense of his reputation (I know I see him as rather juvenile and childish, especially after reading some of his emails on the subject) and his technical understanding (he brushes aside all technical explanations as to why his code should not be included with a, "Yes, yes, but you're full of crap - it's a vendetta against me by IBM and SGI and Linus Torvalds and Stephen Tweedie and the men in black and the grey aliens and Magneto and Kermit the Frog and Big Bird and the Powerpuff Girls and...").

    If there is money to be made by having a journalled file system, it can already be made without having his file system in the mainline kernel source. But rather than do that, Reiser spends his time fighting a losing battle for a meaningless accomplishment while losing the respect and credibility of the very people he wants to impress. Rather than defend his product on its technical merits and features, he spends his energies on a battle he very clearly won't win.

    It sounds to me like the monetary reward Hans Reiser so clearly expects is not coming because it's his own damn fault, not because his code doesn't get in the standard kernel source.
    --
  • Since when does Linux need anything? It works fine for me; all my favorite software runs on it. Therefore I can say that Linux categorically does not need anything, based on my experience.

    "It works for me so it doesn't need anything else", I suppose if you have a couple gig of ram you don't NEED virtual memory either, or if you have an x86 processor you don't NEED an alpha or sparc port. Not to mention modules, who needs modules if all your hardware is initialized by the BIOS?

    The fact is pretty much every other unix has some type of journaling filesystem where linux simply does not. Linux is mostly used as a server OS and therefore needs server OS features. Journaling isn't "wiz-bang" anymore, it's essential, even for the home user. An fsck on a 30 gig drive will show you that.

    -- iCEBaLM
  • Keep in mind that JFS does not protect your data integrity (it only journals file system structures), and it also doesn't protect you against hardware failure, bugs, or misconfiguration.

    So, for anything that is critical or costly, you need redundant hardware and replication anyway. Once you have that, you are already protected against downtime from running an fsck should the need ever arise. There is no need to slow down your (replicated) systems additionally with JFS for a very marginal benefit.

  • by jetson123 ( 13128 ) on Thursday October 05, 2000 @06:36AM (#729965)
    I was using JFS under AIX for several years. I can't imagine why people would want to run it. It doesn't give you much more data security (only file system structure is journalled), and you pay a heavy price in terms of performance. In fact, running Linux and AIX side-by-side for several years, Linux with ext2 on a low-end IDE drive not only greatly out-performed AIX on a high-end workstation and SCSI drive, AIX even lost a file system during a crash.

    Fsck on ext2 is pretty fast, crashes are very rare for server systems, and servers require regular backups anyway. It is more rational to run integrity checks in batch mode when necessary than to pay overhead on every file system access to deal with the possibility that the machine might crash at any moment. I think JFS (and its companion, LVM) are simply not good engineering tradeoffs for most (all?) applications.

  • by josepha48 ( 13953 ) on Thursday October 05, 2000 @06:37AM (#729966) Journal
    .. including everything.

    It seems these days that everything is getting into the Linux kernel. While this is great, I imagine there are going to be some newbiew that are thinking which should I choose, which is better. The answer is more of what are your needs.

    I wonder how distros are going to handle this. If a distro includes only support for one, then a user will have to choose the distro based on the fs he wants, if they include all the user then has to choose which one he wants to use.

    Since this is compiled into the kernel can you use more than one journaling file sytem at a time?
    Can I have one partition using ext3fs, one using reiser, and one using IBM's jfs, and one using XFS? Not that I'd want to though.

    I don't want a lot, I just want it all!
    Flame away, I have a hose!

  • I heard that somebody was re-porting it to OS/2. How is it going?
    __
  • "You have got a source haven't you?"

    I gave you a good example, isn't that enough? Go get the stopwatch and start browsing!

  • If you think that the 2.4 kernel is 'long overdue' then why don't you volunteer and help out the effort?

    Not allowed to.

    They get to tell you to fix your problem yourself.

    And then ignore your patches.


    ________________________________________
  • by Otis_INF ( 130595 ) on Thursday October 05, 2000 @05:04AM (#729983) Homepage
    I mean: if people stop coding instantly and start designing the stuff for a change, the VFS can get universal for _ALL_ kind of filesystems: plug 'em in and they'll run. Why o why do people start coding their own freaking filesystem without a proper motherlayer first? I know there are rivaling opinions about what the VFS should do and what it shouldn't do but, please... this kind of software development seems to me erm... rather unsuccesful.

    "Hey, the last 9 months I programmed on this superduper filesystem and it will be great for the next Linux Kernel!"

    "Erm.. yeah, great but the VFS layer isn't up to par so we can't use your functionality in the rest of the system anyway"

    Besides that... Linus isn't stupid. He already mentioned a zillion times he wants to end the featurecreep and finish the kernel. Now adding another filesystem will definitely delay the kernel's release BECAUSE of featurecreep, something Linus wants to avoid.

    But.. with a better design of the system internals, this wouldn't have to be necessary: IBM would just add another module and everything would have been fine. ah well...

    Good old.. mr. Tanenbaum ;)
    --

  • by Dios ( 83038 ) on Thursday October 05, 2000 @03:41AM (#729989) Homepage
    Hmm.

    as far as I have read on kernel threads (http://kt.linuxcare.com [linuxcare.com]) there will be no journalling systems in 2.4, initially at least.

    This is kind of an odd announcement especially with the kernel in its current 'slushy' state.. I am guessing that 2.4.X with X>5 before the first journalling systems begin to appear.

  • The port is being done from the OS/2 version and in order to keep compatibility with OS/2 the open source version is allowing a case insensitive mode

    The latest drops however have implemented the case sensitive mode and made it the default.

    For those interested the you can get on the jfs-discussion mailing list by sending email to Majordom@oss.software.ibm.com with the following command in the body of your email message:

    subscribe jfs-discussion
    Information on IBM's OSS projects can be obtained from http://oss.software.ibm.c om/ developerworks/opensource/ [ibm.com]

    And as far as the "buffet" of Filesystems goes: choice is always good, lots of choices even better.

  • There was an undelete mechanism during the 2.1.x development system. I played with it for a few days.

    When an inode was unlinked it was not really unlinked, but moved to a file whose name began with a dot. Once enough of these deleted files had accumulated in one directory, subsequent unlinks would really unlink the oldest.

    When the d-entry code came in, this disappeared.

    • The current JFS for Linux project is, for instance, still case insensitive. Hardly an acceptable situation for a UNIX filesystem, but hopefully one that can be fixed.
    actually, if you'd go look [ibm.com], you'd see that as of release 13 of JFS (September 29, 2000) is now case sensitive. But I suppose you can't be bothered to research something before you flame about it.

    Having been a contractor at IBM in Austin, and having spolen personally with some of the AIX developers and their managers, I can state that IBM groks open source. (at least the departments I've had contact with did) IBM's JFS may not be the most mature journalling filesystem for linux out there, they aren't claiming to be, but it is another one. When it's all done, we'll probably have a couple of jfs's left (one that only journals meta-data, and one that journals both data and meta-data) and they'll likely have incorporated the best pieces of the jfs's that have fallen aside. Isn't that what it's all about? So all you people who spread doom and gloom about IBM and open source, just stop it. We're sick of it ok?? What about Sun?? They've pissed on the open source community and yet what of them? Oh, that's right... they're not IBM. Evil big blue brother IBM. get off your high horses and accept that they're doing something good for you.
  • Thanks, Hans.

    <Grin>.

    But seriously: that AC was right. ReiserFS is more than just a journaling filesystem, and it's rock solid. I started using it on my laptop as a patch applied to my 2.2 kernel. I had also been using Suse's USB backport patch, but it was crashing my laptop (some kind of interaction with APM), and I got tired of waiting for 8GB of ext2fs partitions to fsck.

    Under normal operation, the ReiserFS is fast and reliable. In fact, most of the time I forget that there's anything special about the filesystem. In recovering from a crash, though, the reisers really shine -- they recover nearly instantly. Only once have I ever lost data: my battery ran down, and a file I had been editing was empty after I restarted the laptop. But even this may not have been the fault of the filesystem; the laptop may have powered down at just the wrong time during the file-save cycle, just after ftruncate, but before any data had been written.

    Anyway, journaling filesystems are not magic -- they can lose data. Read that again: they can lose data, just like any filesystem. They just recover much faster because they guarantee the integrity of the metadata.

    Bottom line: ReiserFS gets two thumbs up. Highly recommended if you're not afraid of patching your kernel.

    --Jim
  • The article states that case sensitivity has yet to be implemented. I'm not sure I understand...

    I would have thought that case insensitivity would be the sort of thing you have to put effort into developing -- and that case sensitivity is something that "just happens"? ... and AIX's JFS is already case sensitive, so what's changed?

    Can anyone explain what I'm missing?
    --
  • by Troed ( 102527 ) on Thursday October 05, 2000 @03:45AM (#730016) Homepage Journal
    Can this be combined with a secure filesystem? We're talking so much about securing our machines, but very few users run them with a secure filesystem. Myself I only use e4m with mapped drives on my Windows machine, which isn't good enough (doesn't encrypt web caches etc). Any new file system, whether it's journaling or not, should have built in security if it can't be done on top of it.

    (When I log out of my machine, I want to _know_ that the contents on it can be reached by myself only - no matter what any repairman, hacker or ... police, does)

  • Interesting. I was able to get the page without problem. In case you don't get the chance, JFS stands for Journeled File System.

    Simply put, Journeling File Systems keep track of where data is placed on a file system. Whenever you move files, add or remove files, etc., the file system knows where those files go (by writing to a "journel").

    The good thing about this is that when a computer goes down, the filesystem can use the journel to put things back in order. Otherwise, it must go through sector by sector and find the data that it has "lost". This causes filesystem checking software (like fsck) to work dramatically faster.

    The bad thing about this is that these types of file systems can take a performance hit if a lot of disk IO goes on. I hope you get a chance to read the article, as it addresses some of these issues.

    --
    Mando
  • by Anonymous Coward
    This is like the promise of that RedHat boss who claimed last year or the beginning of this that 2.4 is going to be released in march 2000...
    Linus just isn't going to let those journalled filesystems in until the VFS layer has been worked out so that the different fs's can share functionality instead of doing everything on their own.
  • by Anonymous Coward on Thursday October 05, 2000 @03:49AM (#730021)
    If you think that the 2.4 kernel is 'long overdue' then why don't you volunteer and help out the effort? That's the benefit of Open Source Software, whenever someone makes a negative statement about the software, then one gets to tell them to 'work on it themselves'. it doesn't matter if you are busy, employed, don't know C or C++, or that the person telling you this has never even seen the kernel source code. They get to tell you to fix your problem yourself.

    Isn't Open Source wonderful?

  • Native S/390 support and a journaled file system. I think we are seeing the beginnings of a new breed of Linux.
  • Actually I used JFS for a couple of years (around '95) and quite liked it. The filesystem is dynamically expandable so you can change your partitioning and add space to a JFS filesystem without having to boot single user, backup, delete, create, restore. Of course you have to have a dynamic partitioning system to do this, but that is why LVM is so useful.

    And really I can't imagine any filesystem that could be absolute proof against anecdotal failures like the one you experienced; there are after all head crashes and the like.

    That said, I do think that Linux is much easier to administrate than AIX (for SunOS junkies Linux is really easy to pick up, and the automation utilities don't mess with the configuration files the same way that OSF/1 decided to do.)

    But even just for personal use I think that dynamically expanding the filesystem would be worth the installation effort, and for people living in the San Francisco bay area where we've been having rolling blackouts over the summer, avoiding fsck's every hot day is worth a minor hit on directory writes.

  • There seems to be a lot of politics surrounding ReiserFS's exclusion from the mainstream kernel.

    So I'm not the only one getting that impression. It seems to me, too, that he's getting the runaround. "Oh, I'm sorry Hans, but we can't possibly put your filesystem into the kernel until we get this fancy new VM layer finished." Never mind that reiserfs has been working beautifully in the 2.2 kernel for a long time, and probably works as well in 2.4 (I haven't tried it).

    Hans appears to be trying to play by the rules, but the system seems to be rigged against him. But if Hans dares to express some frustration with the situation, he gets flamed hairless by the kernel insiders. I'm sure the apologists for the current process can drum up many valid technical reasons for the delay in integrating reiserfs into the mainstream, but to this outsider it looks like Hans Reiser is getting abused by the system.

    What's really sad is that it's the many casual linux users -- people who would use reiserfs if it were in the kernel, but are afraid of kernel patches, or don't have the time to figure them out -- who are paying the price.

    Comments? Flames?

    --Jim
  • Well, duh, that's what tracert and telnet are for!
  • Then why don't you just do a tracert to winamp.com and slashdot.org and subtract the ping times from the total time? You did graduate from elementary school, didn't you?
  • He says the editor may have just emptied the disk file prior to writing the new data. If it crashed in between, the journalled data would have been correctly empty. Nothing any journal file system can do about that, unless it also includes version control in some form.

    --
  • I couldn't mount dos partitions anymore...

    You screwed up. My laptop can mount ext2, reiserfs, and vfat (dos + win95 long file naming) with no problem. I also have the international crypto patch and can mount encrypted loopback filesystems. A pretty exotic mix, but it all works beautifully, at least on a stable 2.2 kernel.

    --Jim

To be is to program.

Working...