Weekly Edition
Return to the Press page
| |||||||||||||
Linux Summit will preview new advanced file system (SearchEnterpriseLinux.com)
SearchEnterpriseLinux.com previews presentations by Ted Ts'o and Chris Mason at the Linux Foundation End User Collaboration Summit that starts on October 13. "But changing the file system to fix the scalability and functional limitations of ext3, the default file system in many popular Linux distributions, requires a significant education outreach. Because the consequences of data loss are so severe, data center managers are reluctant to trust their data to new file systems, Ts'o said. New-system information needs to be shared well ahead of time, including a roadmap of coming features so IT professionals know what to expect, he said. That's where the Linux Foundation's event hopes to make inroads."
(Log in to post comments)
Question Posted Oct 13, 2008 18:31 UTC (Mon) by kragil (subscriber, #34373) [Link]
Btrfs and Ext4 are great. No doubt.
But what about Tux3? From what I have seen the kernel developers seem to largely ignore it.(No answers to the postings about it.)
Why is that?
Question Posted Oct 13, 2008 18:42 UTC (Mon) by corbet (editor, #1) [Link] The kernel developers haven't seen tux3. When the code is posted for their consideration, they will consider it.
Question Posted Oct 13, 2008 19:55 UTC (Mon) by kragil (subscriber, #34373) [Link]
Ok I see, thank you.
It is all available though ( http://tux3.org/ ). But it is bitbucket/Mercurial hosted. That might explain a lot ;)
Question Posted Oct 13, 2008 20:28 UTC (Mon) by rahulsundaram (subscriber, #21946) [Link]
Usually, hosting it in a external repository somewhere just doesn't cut it. It needs to be posted to LKML as a series of patches allowing specific feedback and then respond with more iterations till it gets into a acceptable state.
Question Posted Oct 13, 2008 21:57 UTC (Mon) by njs (guest, #40338) [Link]
btrfs hasn't been posted for inclusion either, though, has it?
Question Posted Oct 13, 2008 23:04 UTC (Mon) by kragil (subscriber, #34373) [Link]
Interesting thread (especially Teds mail). So Btrfs is pushed by a lot of companies... And the masterplan is to go from ext3 to ext4 and then to btrfs.
OK, but I still think it is good to have a choice. Having a good selection of feature rich filesystems will not hurt. Distros pick their favorites anyways.
Question Posted Oct 14, 2008 4:16 UTC (Tue) by jamesh (guest, #1159) [Link]
Note that most distributions have settled down on ext3 at the moment. I don't see many trying to differentiate themselves on the basis of the default file system choice.
That might change post-ext4, but then again distributions might continue to cooperate. If we assume that most distributions switch to ext4 in the short term (this seems likely due to the low cost of switching), that leaves a decent amount of time for both btrfs and tux3 to try and prove themselves.
Question Posted Oct 14, 2008 17:46 UTC (Tue) by drag (subscriber, #31333) [Link]
Ext3 vs Ext4 is going to be a hard transition for distributions because it's not obvious that Ext4 is going to have lots of improvements that will have a substantial impact on end user experience. Does 'now have extents' really going to make Openoffice start faster or make my system more 'crash resistant'?
Btrfs, on the other hand, has that really 'wow neat features' aspect to it. The same sort of attractive stuff that makes people lust after ZFS without ever actually touching it or knowing that much about it or whether or not it will eat their data or run with any sort of acceptable performance.
It seems pretty likely that, at this point, people would end up supporting Ext4 for the minority of people that specifically request it and leapfrog it by making Btrfs the real successor to Ext3 (by way of the default FS selected by their installers).
Question Posted Oct 15, 2008 0:13 UTC (Wed) by qg6te2 (guest, #52587) [Link] Significantly shorter fsck times for Ext4 is a compelling reason by itself. Reduced fragmentation and being able to handle more files as well as larger hard drives is yet more reasons. As good as btrfs is going to be, it is years away from completion ("production ready" is being touted for 2012), vs Ext4 that is just around the corner.
Ted's email Posted Oct 16, 2008 13:32 UTC (Thu) by engla (guest, #47454) [Link] Ted's email on ext4/btrfs is indeed very interesting, even carrying old but not previously known news. Thank you! http://thread.gmane.org/gmane.linux.file-systems/26246/focus=26317Snippet As far as btrfs is concerned, one of the things that you may not know is that about a year ago (on November 12-13, 2007), a small group key filesystem developers, that included engineers employed by HP, Oracle, IBM, Intel, HP, and Red Hat, and whose experience included working with a large number of filesystems: ext2, ext4, ext4, ocfs2, lustre, btrfs, advfs, reiserfs, and xfs came together for a two day "next generation filesystem" (NGFS) workshop. At the end of the that workshop, there was unaminous agreement (including from yours truly) that (a) Linux needed a next generation filesystem to be competitive, (b) Chris Mason's btrfs (with some changes/enhancements discussed during the workshop) was the best long-term solution for NGFS, and (c) because creating a new enterprise filesystem always takes longer than people expect, and even then, it takes a while for enterprise users to trust a new filesystem for their most critical data, ext4 in the next generation of filesystems was needed as the bridge to the NGFS.
backwards compatible?? Posted Oct 13, 2008 23:51 UTC (Mon) by dlang (✭ supporter ✭, #313) [Link]
that's the first time I've ever seen anyone say that ext4 would be backwards compatible to ext3.
useing much of the same codebase I believe (it's mostly the same developers), but everything that I have read about ext4 indicates that there is no chance at of of being backwards compatible with ext3
backwards compatible?? Posted Oct 14, 2008 0:13 UTC (Tue) by qg6te2 (guest, #52587) [Link] Ext4 is backwards-compatible in the sense that it can mount Ext3 partitions. Ext3, on the other hand, is not fully forwards-compatible with Ext4. The forwards-incompatability arises when an Ext3 partition is converted to use the extents feature of Ext4.
backwards compatible?? Posted Oct 14, 2008 0:15 UTC (Tue) by csamuel (✭ supporter ✭, #2624) [Link]
Note that this is the default when an ext3 partition is mounted as ext4!
FOSS and multiple competing commercial interests Posted Oct 14, 2008 8:22 UTC (Tue) by kripkenstein (subscriber, #43281) [Link]
This is quite interesting to me from a licensing standpoint. It appears, if I read this correctly, that a group of large commercial interests are banding together to work on btrfs, as the cost of doing so separately is too high, and they have decided on the GPL license for that?
First issue that I am curious about: I am aware that btrfs must be GPL-*compatible* to be usable in Linux, that's a given. But it could have been BSD, zlib, or something else permissive. That would have allowed btrfs to be used in OpenSolaris, the BSDs (,Windows) and so forth. Is this intentional, to disallow this?
Another issue, is this not a fairly unique occurrence, that several (competing, in general) business interests band together to found a new open source project? What I mean is, Linux and other major projects with contributions from several sources generally grew 'organically' - business interests joined the project over time. But *new* projects generally have a single clear owner, and here that is not the case? Interesting, if so.
FOSS and multiple competing commercial interests Posted Oct 14, 2008 8:57 UTC (Tue) by tao (subscriber, #17563) [Link]
Maybe the licensing choice is a direct reaction to the anti-GPL licensing choice by Sun?
GPL is the only popular license which can be used here Posted Oct 14, 2008 10:06 UTC (Tue) by khim (subscriber, #9252) [Link] BSD or zlib license will quickly lead to the new round of the unix wars - and this is certainly what the creators of btrfs don't like to see. CDDL is incompatible with GPL and so is unusable. *BSD systems can use btrfs just fine (they support ext2, after all - and it's GPL-licensed), Windows can use it too (unlike linux they have stable ABI for FS drivers and it'll work as GPL boundary), so the only loss is Open Solaris - and I think creators was sure it's a plus :-)
GPL is the only popular license which can be used here Posted Oct 14, 2008 10:14 UTC (Tue) by kripkenstein (subscriber, #43281) [Link]
> *BSD systems can use btrfs just fine (they support ext2, after all - and it's GPL-licensed)
I think I don't understand something here. If ext2 (and ext3) and GPL licensed, how can a BSD operating system use them? As I understand it, that would require that all other code in the OS be GPL-compatible, as effectively the OS as a whole is now under the GPL. The BSD license is in fact GPL-compatible, is that how this works?
The BSD license is in fact GPL-compatible, is that how this works? Posted Oct 14, 2008 11:50 UTC (Tue) by khim (subscriber, #9252) [Link] Yup. The whole kernel of *BSD OS can be relicensed under GPL so it can include GPLed filesystem. Nobody will do this in practice, of course, but the fact that you CAN gives you the right. Proprietary forks can not use btrfs, of course...
The BSD license is in fact GPL-compatible, is that how this works? Posted Oct 14, 2008 11:59 UTC (Tue) by kripkenstein (subscriber, #43281) [Link]
Thanks for the clarification.
It seems then that BSD kernels can *not* include both GPL and GPL-incompatible code, like Apache (which IIRC is incompatible with GPL2) and CDDL? In other words, in practice you can use BSD with either ZFS (CDDL) or ext3 (GPL), but not both at the 'same time', in some appropriate sense?
If so then that puts the BSDs in an interesting (precarious, in a way?) position.
Kernel? No. CD? Yes. Posted Oct 14, 2008 13:28 UTC (Tue) by khim (subscriber, #9252) [Link] It seems then that BSD kernels can *not* include both GPL and GPL-incompatible code, like Apache (which IIRC is incompatible with GPL2) and CDDL? Yup. If we are talking about single binary blob. In other words, in practice you can use BSD with either ZFS (CDDL) or ext3 (GPL), but not both at the 'same time', in some appropriate sense? You certainly can distribute CD with *BSD kernel, GPLed ext3 and CDDLed ZFS. "Mere aggregation" clause makes this distribution possible: ZFS and ext3 modules are usable without single byte of code from other module thus they are not the "part of the same program" and kernel is compatible with ext3 module and with ZFS module. Can you load both modules simultaneously? This question will need court decision, but I doubt ext3fs developers will sue *BSD users over this problem so we'll probably never know... If so then that puts the BSDs in an interesting (precarious, in a way?) position. Actually just their users not the developers or distributors... And as it's highly unlikely that ext3fs developers will sue *BSD users... I doubt we'll ever know for certain if it's legal or not...
Kernel? No. CD? Yes. Posted Oct 14, 2008 13:31 UTC (Tue) by kripkenstein (subscriber, #43281) [Link]
Ok, that clears it up for me, thanks.
Kernel? No. CD? Yes. Posted Oct 14, 2008 14:08 UTC (Tue) by salimma (subscriber, #34460) [Link]
Same is true for ZFS-on-Linux, though in that situation, our assurance that Sun won't sue comes from an individual Sun ZFS developer, and so does not carry much weight.
Kernel? No. CD? Yes. Posted Oct 14, 2008 14:33 UTC (Tue) by paulj (subscriber, #341) [Link]
Same is true for ZFS-on-Linux, though in that situation, our assurance that Sun won't sue comes from an individual Sun ZFS developer, and so does not carry much weight. FUD.. Your assurance that Sun wouldn't sue comes from the CDDL licence. The prevailing wisdom on LWN is that distributors of a hypothethical ZFS-on-Linux would have to be fearful of Linux copyright holders sueing..
Kernel? No. CD? Yes. Posted Oct 14, 2008 17:59 UTC (Tue) by drag (subscriber, #31333) [Link]
According to the CDDL license binding it to Linux (which is under the GPL) is illegal.
That's all there really is to it. Even if a Sun developer has a 'promise not to sue' attitude.. Which are you going to beleive?
A) The guy's blog
?
So it really doesn't fundamentally matter what some guy says or what the company releases in the press release. It doesn't matter if you or I beleive them or any such thing. The CDDL is incompatible by design, that is to say: "on purpose", and if Sun wants to make it legal to have ZFS on Linux then it's up to them to modify the CDDL or dual license it to allow that to happen.
That is how you know they are serious or not. Everything else is just hand waving.
They could say that from now on the sky is going to be green and purple. And unless the sky eventually changes color you know that they are probably full of crap, no matter how much they beleive it themselves.
As far as Linux goes... ZFS is a file system. File systems are very hard to do and get stable. No Linux kernel hacker would be willing to touch ZFS with a ten foot pole (since the original copyright holder has made it obvious through the bundled license that it's not to be used with Linux) and thus it has ZERO chance of being incorporated into Linux unless somebody takes the time to rewrite a ZFS implimentation completely from scratch. And if the File system doesn't get officially incorporated into Linux then it'll end up like ReiserFsv4. Nobody else will care, nobody would use it for anything important if they have half a brain.
Kernel? No. CD? Yes. Posted Oct 14, 2008 22:25 UTC (Tue) by paulj (subscriber, #341) [Link] According to the CDDL license binding it to Linux (which is under the GPL) is illegal. Can you point out the problem clause? That'd be news to a good few people. You're asking us now to take your opinion on authority, while at the same time arguing against the authority of a well-respected Sun developer. That doesn't make sense. Next, you argue from an assumption that Sun wants, nay must to donate its code to Linux. This assumption is clearly flawed, disappointing as that may be to you. I won't speculate much here, but I think Bryan Cantrill has commented on LWN articles before about what Sun wanted. I'll just note that Sun ZFS and DTrace is shipping just fine with other OSes (proprietary and free). The code is out there, it's clearly free to re-use. It's not Suns' job to have to convince one group of people that they should find a way to agree to not sue each other for distributing CDDL code.. NB: I enjoy reading your comments on LWN, and apologise in advance for any possible offence. ;)
Kernel? No. CD? Yes. Posted Oct 14, 2008 22:37 UTC (Tue) by paulj (subscriber, #341) [Link]
Oh, reminder: I work for Sun.
Kernel? No. CD? Yes. Posted Oct 15, 2008 22:42 UTC (Wed) by drag (subscriber, #31333) [Link]
> NB: I enjoy reading your comments on LWN, and apologise in advance for any possible offence. ;)
Oh, don't worry about that. I try to get out there and say things so that people challenge me and try to correct my misconceptions. So I am much more worried about offending other people, then worrying about them offending me. Since I depend on them.
> Can you point out the problem clause? That'd be news to a good few people.
I think the biggest one is the explicite patent license stuff in second 6 and mentioned other places on the license. This sort of thing is a extra requirement and thus violates the GPL notions of 'no additional restrictions'
I was hoping that the GPLv3 would be more compatible, but as far as the Linux kernel it's going to be GPLv2 for pretty much the rest of enternity at this point.
The CDDL is a fine license otherwise. Much more well suited for what Sun needs.
It's essentially a cleaned up version of the MPL and the Mozilla stuff has to be tripled license to remain GPL compatible.
Different parts of Sun seem to agree with me. Netbeans, for example, is dual licensed under CDDL and GPLv2+classpath exception. Could be just political theater (which isn't bad), but I donno.
I think that Linux developers are generally not going to touch ZFS unless they get a specific nod from Sun itself, probably in the form of a GPL exception or dual license, for the ZFS stuff.
"Sun wants" Posted Oct 15, 2008 10:00 UTC (Wed) by tialaramex (subscriber, #21167) [Link]
Next, you argue from an assumption that Sun wants, nay must to donate its code to Linux. This assumption is clearly flawed, disappointing as that may be to you. I won't speculate much here, but I think Bryan Cantrill has commented on LWN articles before about what Sun wanted. I'll just note that Sun ZFS and DTrace is shipping just fine with other OSes (proprietary and free).
There is no such assumption, nor an argument based on it. Sun hasn't been capable of the single-mindedness you're implying for at least a decade. Some elements at Sun still think it is 1995 and are doing the "You can't take Linux seriously" line, which even Microsoft grew out of eventually. Some elements want Sun to repeat its earlier transformation and put Solaris out to pasture in favour of new code with a compatibility layer. It should be no surprise that people paid to maintain Solaris don't like /that/ idea. Some elements want Sun to be a hardware company and jettison most of the software, others think the hardware is just a boat anchor and Sun ought to be hardware-agnostic. Clearly some elements at Sun would be delighted to see ZFS everywhere, while others think (probably wrongly) that it's a secret sauce worth keeping from the "competition" whoever that is today.
We're used to this in Linux. Linux doesn't "want" anything in particular either. But it does seem laughable when individual Sun employees turn up to tell us what "Sun" does or does not want and it just reflects their own personal feelings. I've worked for a tiny company with maybe twelve individual employees, and still you would get a very different idea of what the company's direction was from the CEO than from Sales or Engineering people.
If you want an idea of the Linux experience of individual smart people turning up and making promises about license conditions, go look at the Larry McVoy/ Bitkeeper history. That's what Bryan's word is worth to us. Citing other projects as willing to include ZFS neither answers the issue of its license nor even tells us what Sun's real intentions are.
"Sun wants" Posted Oct 15, 2008 12:47 UTC (Wed) by paulj (subscriber, #341) [Link] There is no such assumption, nor an argument based on it. I guess I misunderstood drag then, where he wrote:
if Sun wants to make it legal to have ZFS on Linux then it's up to them to.... That is how you know they are serious or not. As for corporate single-mindedness, few non-trivial corporations are capable of it. That said they most definitely are capable of making single-minded, executive decisions - and Suns' mind in that respect wrt open-sourcing Solaris is very clear. That decision was made based on some weighting of the corporation's collective wisdom. Various weighty Sun people, from executives to notable engineers have given their view, in various places. Their opinions are not Suns' per se, but they are most definately a significant subset.. To go further would be to philosophise way too much about corporate structures, so I'll stop. ;) As for BitKeeper - When did LMcVoy ever open-source BitKeeper? What data of Linux has Sun locked in a proprietary format?? This seems a very flawed analogy (and please, let's try to avoid arguing too much by analogy). I'm not asking for LWN readers to start loving Sun (though, a lot of Sun engineering work does go to benefit Unix generally, including Linux). However arguments based around FUD, logical fallacies and false-equivalences seem quite below the normal, high standards of commentators here. Would grudging respect be too much to ask for? :) (you may assume I'm inclined to make similar arguments at work, if the need arises, not that my opinion carries much of a weighting ;) ).
"Sun wants" Posted Oct 15, 2008 13:02 UTC (Wed) by paulj (subscriber, #341) [Link]
Ok, I think I see what you're trying to say, and it's already been dealt with in my earlier replies, but lets try again: You don't need Bryan Cantrill's word, cause the code is published under the CDDL. You can just ask your own lawyer. Where do you get this FUD from btw? I know Jon published an "Beware the evils of Sun!" piece here, but his claims about CDDL/GPL compatibility there were highly specific to GPL projects where contributors mistrust each other so much they are incapable of agreeing not to sue each other for incorporating and distributing 3rd-pary free software.. ;)
Quite succinct description of Linux kernel, yes... Posted Oct 15, 2008 13:48 UTC (Wed) by khim (subscriber, #9252) [Link] Yahoo! Exactly! That's definition of linux kernel! Fact 1: SCO is
kernel
contributor. Do you STILL want to say that linux is not an GPL project where contributors mistrust each other so much they are incapable of agreeing not to sue each other for incorporating and distributing 3rd-pary free software.. ;) Fact is: when you include "big boyz" of software world (and list of kernel contributors read like "who is who in the world of IT" - only Microsoft and Rambus are not included... yet) in a single project... litigation is not out of the question. And how the hell can you even think what you wrote when SCO is not yet liquidated... it's frankly beyond my understanding...
Quite succinct description of Linux kernel, yes... Posted Oct 15, 2008 14:02 UTC (Wed) by paulj (subscriber, #341) [Link]
I am responding specifically to the accusation that Sun-copyrighted CDDL code being incorporated with GPLed code could lead to those involved in the GPLed code being sued by Sun.
That's FUD..
So your reply seems misdirected.
That's not ACCUSATION, that's POSSIBILITY... Posted Oct 15, 2008 14:26 UTC (Wed) by khim (subscriber, #9252) [Link] I am responding specifically to the accusation that Sun- copyrighted CDDL code being incorporated with GPLed code could lead to those involved in the GPLed code being sued by Sun. What kind of response is this? Could Sun sue these people? Absolutely: Sun is contributor to Linux Kernel and so will have the right to sue. Will Sun sue such contributors? Who knows? When Caldera sponsored SMP patches to Linux is was also "quote obvious" it'll not sue other contributors... and it turned out to be not true. Corporations are not people - you can not trust them! CEO come and go, shareholders can overturn CEO decisions, etc. Only written binding contracts can be trusted. That's FUD.. Yes, it's FUD, but it's not groundless FUD. This scenario is quite real - no matter how unlikely it looks today. ZFS adoption can affect millions (perhaps billions) of people. To base their wellfare on faith in future Sun's decitions... looks reckless. It all comes down to the question: do you trust FSF lawers more then you trust Sun's developers where licensing schemes are concerned? And sure as hell I'll trust FSF's lawers more then Sun's developers. And not because I don't trust Sun's developers.
That's not ACCUSATION, that's POSSIBILITY... Posted Oct 15, 2008 15:19 UTC (Wed) by paulj (subscriber, #341) [Link]
So Sun would be the one sueing, but not using the CDDL, but for GPL infringement? Even where the GPL copyright holders have agreed to allow CDDL code? I have to say I havn't thought about that one, and I presume the judge won't much either before throwing out such a suit...
Whether its practical for Linux copyright holders to reach such an agreement, I've no idea. However, for many other projects (GPL, BSD or otherwise), such agreement is easily achievable - and on that count at least some of the CDDL/GPL "Sun could sue you!" claims are just pure FUD.
Anyway.. Let's call it a day now.
Exception for CDDL? Possible yet unlikely... Posted Oct 15, 2008 17:16 UTC (Wed) by khim (subscriber, #9252) [Link] So Sun would be the one sueing, but not using the CDDL, but for GPL infringement? In a word - yes. I have to say I havn't thought about that one, and I presume the judge won't much either before throwing out such a suit... Why not? It's well-known fact that GPL and CDDL are like oil and water. And it's also well-known fact that Linux kernel is licensed in such a way that any "exceptions" are not easy to add (on purpose BTW). So it does not look like a stretch that Sun have chosen CDDL for ZFS to keep it out of Linux kernel - making both free, but separate pieces of code. If this was design all along (and it certainly looks like this from outside) then why do you think Sun will not sue the trespassers? Especially if it has the right to do so? Even where the GPL copyright holders have agreed to allow CDDL code? Sun is one of these holders, you know. Sun, too, will be forced to sign the paper which says "as special exception you can, blah, blah, blah". Along with Broadcom, Dell, IBM, Intel and countless others. And if Sun wanted such an outcome it always had other option: just dual-license the thing under CDDL and GPL (like Mozilla did). Because Sun does not do what is simple (GPL already has some kind of approval from Sun's lawers - it's license of OpenOffice.org, after all) and instead preaches strange legal theories on blogs and forums the general consensus is: "either Sun's management does not want to have ZFS in Linux at all or it wants to use it as a trap". What Sun's developers think is not very relevant here. Neither choice very appealing but most people thing that first opinion is closer to the truth (because second is much uglier). Whether its practical for Linux copyright holders to reach such an agreement, I've no idea. Possible but impractical. It's more-or-less the same process as a switch to GPLv3 and it was investigated quite thoroughly. If Linux deverloper wanted ZFS really badly they can always start adding such exception, ask Sun to sign it too (and get approval or rejection in which case they'll be forced to rewrite parts of kernel), etc. I'm afraid it takes at least few years and by then btrfs will be ready to do what ZFS does now. Yes, ZFS is great but it's not THAT great... However, for many other projects (GPL, BSD or otherwise), such agreement is easily achievable - and on that count at least some of the CDDL/GPL "Sun could sue you!" claims are just pure FUD. What other GPLed kernel are out there to use ZFS? Other CDDL projects can be usable (and will be usable, I'm sure). Just not ZFS. I think dtrace can be usable, for example (you need hooks in kernel, true - but these must be rewritten anyway and userspace<->kernelspace barrier was always considered as GPL barrier so CDDLed userspace tools can be used).
Exception for CDDL? Possible yet unlikely... Posted Oct 16, 2008 8:31 UTC (Thu) by paulj (subscriber, #341) [Link]
And if Sun wanted such an outcome it always had other option: just dual-license the thing under CDDL and GPL (like Mozilla did). This makes sense only if you ignore the various, quite sensible reasons given by several Sun notables as to why the GPL and/or dual-licensing would not have worked for Solaris. Dual-licensing: I'm reasonably sure at least one very well-respected engineer would have argued strongly against dual-licensing. It can lead to licence-forks. I.e. dual-licensing is worse than choosing the inferior of the two (incompatible) licences. GPL: Is very incompatible with other licences (as we know from this thread). I.e. it would have made it difficult for Sun and others particularly to ship things like non-GPL (e.g. proprietary) drivers, filesystems and volume-managers with Solaris. Further, the GPLv2 does not really address patent issues, and Sun is a regular target for patent suits. So yes, Sun could have chosen the GPL, but instead chose the CDDL simply to frustrate Linux users - this makes perfect sense if you just block out all the other, practical reasons that relate to Sun and its needs for Solaris. Really, the problem here is that Linux contributors don't sign contributors agreements. Rather than figure out how to fix that and regain some kind of collectively-empowered, executive control, you instead direct your ire at Sun...
And these people are talking about FUD? Posted Oct 16, 2008 13:55 UTC (Thu) by khim (subscriber, #9252) [Link] This makes sense only if you ignore the various, quite sensible reasons given by several Sun notables as to why the GPL and/or dual- licensing would not have worked for Solaris. I'm yet to see sensible reasons. Dual-licensing: I'm reasonably sure at least one very well- respected engineer would have argued strongly against dual-licensing. It can lead to licence-forks. I.e. dual-licensing is worse than choosing the inferior of the two (incompatible) licences. Now we have interesting dilemma: position of at least one very well- respected engineer (in Sun, I presume?) vs position of thousands of Linux developers. Hmm... Who to choose. Sorry, Sun's engeeners are revered - but not to this degree. How many projects can you name where dual-licensing actually led to fork? There are hundreds, may be thousands dual-licensed projects (if you count not just huge projects like Firefox but smaller ones like bunch of CPAN addons). I never seen situation where addition of the license option led to fork. Removal? Sure, it can produce forks. XFree86 (changes made it incompatible with GPL, XFree86 is history now), Firefox (in addition to GPL or MPL license you must agree to the logo license and many developers find it quite onerous => we have IceCat and Iceweasel). Addition or options? Never heard of this. How often this happen, really? Can you name at least one sample? Conclusion: FUD. GPL: Is very incompatible with other licences (as we know from this thread). Number of licenses does not matter. I can generate million GPL- compatible and CDDL-incompatible licenses with a simple script in five minutes. Most projects out there are GPL-compatible. Most projects out there are CDDL-incompatible. It was discussed to death. Conclusion: FUD. I.e. it would have made it difficult for Sun and others particularly to ship things like non-GPL (e.g. proprietary) drivers, filesystems and volume-managers with Solaris. Somehow this is not a problem for Java or OpenOffice.org but problem for Solaris? Conclusion: FUD. Further, the GPLv2 does not really address patent issues, and Sun is a regular target for patent suits. This is at least valid reason. But why this is not a problem for Java, then? Java is not important for Java company? don't make me laugh. Conclusion: FUD. So yes, Sun could have chosen the GPL, but instead chose the CDDL simply to frustrate Linux users - this makes perfect sense if you just block out all the other, practical reasons that relate to Sun and its needs for Solaris. No it makes perfect sense if you know for the fact that Sun released other stategically important projects under GPL before releasing Solaris under CDDL. What exactly is different with Solaris and how it's needs differ from Java and OpenOffice.org? I never heard a plausible explanation. Binary extensions, patents, etc, etc - all these problems are equally applicable to Java, OpenOffice.org, and Solaris. I know if one plausible reason and one plausible reason ONLY: CDDL makes it impossible to implaint Solaris's goodies in Linux and so Solaris gets much needed baubles to preach which can not be just included in Linux. Because Sun's JDK and OpenOffice.org are top dogs in relavant markets and Open Solaris is underdog. Quite valid reason and I'm ready to agree that it works just fine - but why all the pusturing and empty screams "you just don't know about Solaris needs! this mythical secret needs are served so much better by CDDL then GPL - but we'll never disclose them! this is not about Linux! really! believe us!". It just does not look convincing. Really, the problem here is that Linux contributors don't sign contributors agreements. Rather than figure out how to fix that and regain some kind of collectively-empowered, executive control, you instead direct your ire at Sun... No, the problem here is position of Sun who puts itself above ALL OTHER Linux contributors. Basically it says: oh, sure you can use ZFS - but only of YOU ALL (Dell, IBM, Intel, etc) agree to our terms. This is called ultimatum: the rules which serve hundreds of companies and thousands of people should be changed just for Sun or else... you can not use ZFS. Linux community considered this ultimatum and more-or-less unanimously decided: ZFS is great but... it's just not worth it. The ire mostly goes not against Sun's ultimatum. It's Sun's right and there are nothing Linux community can do about it (except accept or reject - it rejected it). Sun does not want to see ZFS included in Linux? Fine - we'll not do this. We respect Sun's wish. It's his filesystem Sun and only Sun can decide if he wants to see it in Linux or not. Ire goes against vain tries of Sun engineers to say that it's not an ultimatum and it's goal is not to keep ZFS out of Linux reach. If it walks like a duck, quacks like a duck, looks like a duck... you can be reasonably sure it is a duck. I'm not sure what game Sun is playing here but to say that everyone else must abandon GPL for unknown reason and claim that "it's not an attack" is just dishonest. Ok, there are another explanation: Sun honestly believe that it's more important then ALL OTHER companies involved in Linux. Dell, IBM, Intel, Novell, RedHad... all their combined voices weight less then voice of at least one very well-respected engineer. If that's so then honestly I don't even know a psychiatrist who can even hope to cure such an acute megalomania...
And these people are talking about FUD? Posted Oct 16, 2008 16:51 UTC (Thu) by paulj (subscriber, #341) [Link]
Whether you agree with the engineers opinion or not on dual-licensing is mostly irrelevant - I'm telling you what the perceived wisdom is among some very senior engineers. You don't have to take their view on dual-licensing on authority at all, you just have to understand that that person is well-respected inside the group responsible for Solaris - that's the cultural background. The point being it has nothing to do with Linux.
As for different Sun software being licensed under different licences, a reasonable person might infer that these licensing decisions are taken in close consultation with, if not by, the groups responsible for the software concerned, and that different groups have different needs and even views. I have absolutely no idea why Java or OpenOffice or whatever are licensed the way they are. My perspective is from the Solaris group..
Anyway, you have your view. I've given mine.
Yet another possibility... Posted Oct 16, 2008 14:02 UTC (Thu) by khim (subscriber, #9252) [Link] Oh, it's possible that real reason is even more subtle: since currently change of Linux's license is very-very hard and there are no entity to sue all this fine arsenal of Microsoft's patents is not really usable. It can sue some distributors and produce ire or EU, but it can not change the license. If there was single entity which can relicense Linux then this entity becomes target of Microsoft's attack with non-zero chance of success. While phrase the problem here is that Linux contributors don't sign contributors agreements points in this direction I just don't believe Microsoft will be sooo deep. It's not above dirty tricks but it's not so subtle... It looks more like scenario of RPG...
Kernel? No. CD? Yes. Posted Oct 14, 2008 14:38 UTC (Tue) by paulj (subscriber, #341) [Link]
Khim,
Forgive me in advance if I have misunderstood your point, but it seems you've overlooked something: The GPL doesn't restrict usage. Users are perfectly free to combine GPL works with proprietary ones.
I think this may even be generally true of software licences considered "free" by the FSF (that includes the CDDL) - though I have not checked.
GPL doesn't restrict usage? True. Is it enough? YMMV. Posted Oct 14, 2008 17:21 UTC (Tue) by khim (subscriber, #9252) [Link] GPL says explicitly that it does not restrict usage ("the act of running the Program is not restricted, and the output from the Program is covered only if its contents constitute a work based on the Program") but then you are not just merely "executing the program": you are creating new program in memory which includes three parts: BSD kernel, GPLed Ext3fs driver and CDDLed ZFS driver. Is it still legal? Different juristictions and even different courts can produce different answers here! The only way to know for sure is via court - and then you'll know the answer for one particular case. Recently Blizzard won similar case, but is it similar enough? Questions, questions, questions...
GPL doesn't restrict usage? True. Is it enough? YMMV. Posted Oct 14, 2008 18:03 UTC (Tue) by drag (subscriber, #31333) [Link]
GPL only matters in distribution. It's the purpose of the GPL. That's it, it's by design. You can see the word 'distribution' repeated many times.
It's not just making copies, it's distributing those copies to other people.
If you do not plan on redistributing any GPL software then you are under no obligation to provide the source code.
Don't mix you desires with facts! Posted Oct 15, 2008 13:21 UTC (Wed) by khim (subscriber, #9252) [Link] GPL only matters in distribution And this is because GPL says ... in part ... . Fill the blanks, please. It's the purpose of the GPL. That's it, it's by design. In your head - may be. Not in real life. These restrictions translate to certain responsibilities for you if you distribute copies of the software, or IF YOU MODIFY IT (emphasis mine). You may not copy, MODIFY, sublicense, or distribute the Program except as expressly provided under this License (emphasis mine again). Does it look like "only distribution does matter" type of license to you? You can see the word 'distribution' repeated many times. In GPLv3 it's propagation. But does not matter: distribution is there, in GPLv2, true, but so is modification. It's restricted, too! Heck: the whole license has a name TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION. The thing which is not restricted is the act of running the Program - and that's all. Can you say without the court decision: combination of works created from BSD kernel, GPLed ext3fs and CDDLed ZFS is "the act of running the program" or "modification of program"? The resulting combination can do things which parts can not do (like copy file from ext3fs-formatted partition to ZFS-formatted one and back)... And these modules surely don't implement some "VFS RFC"...
Don't mix you desires with facts! Posted Oct 15, 2008 13:39 UTC (Wed) by paulj (subscriber, #341) [Link]
The GPL and the modification conditions apply to the preferred form of the sources - not derived binaries (directly at least).
If you got this theory from a lawyer, you need to find a new one, methinks.
GPL does not decide where it's applicable or not - the law does Posted Oct 15, 2008 14:08 UTC (Wed) by khim (subscriber, #9252) [Link] The GPL and the modification conditions apply to the preferred form of the sources - not derived binaries (directly at least). Novel idea. So you claim that I can add any proprietary code to linux as long as I'm not modifying source, but only binary by the way of hex editor? Sorry, if you got this theory from a lawyer, you need to find a new one, methinks. You are mixing cause and consequence. Where and how GPL applies can not and does not written in GPL. Some jurisdictions allow some [limited] form of modifications without license (for example you can reverse-engineer stuff if it's the only way to make it work in Norway - see DVD John), some do not. Read this for example. MDY tried to claim that EULA does not apply to it because it never agreed to its terms, but court decided that they still illegally modified the code in memory - and if it's not allowed without appropriate license then that means GPL should explicitly allow it. It does not. It's pretty clear that GPL does not contain permissions to make such a combination, but do you even NEED such permission? That's the question. It depends on many-many details and can differ from country to country and even from court to court! That's why I never said "yes" about such a thing and never said "no": YMMV. And yes, I've discussed this with actual lawer (it was about combining GPL and CDDL code - though not about linux kernel and ZFS).
GPL does not decide where it's applicable or not - the law does Posted Oct 15, 2008 14:57 UTC (Wed) by paulj (subscriber, #341) [Link] I didn't claim that. :) The binaries are also subject to conditions under the GPL. As for in-ram-linking-as-modification: just look at the modification licence, its framed in terms of conditions that are really only relevant to fixed, distributable copies. It seems very unreasonable, to me, to think the framer had transient modifications in mind. It is then reasonable to think that the alternative is to consider that such transient modifications, made in the course of running, must then be covered by the 'running is unrestricted' permission. That this has been the prevailing wisdom for a long time (e.g. the GPL FAQ, that NVidia are happy to distribute binary-only drivers for users to load with their GPL kernel, etc..) suggests strongly that your lawyer's interpretation is beyond the pale. IANAL, my view is based primarily on assessing the words and actions of those who have (had) significant matters at stake, ICBW, etc.
Prevailing wisdom? May be in your head... Posted Oct 15, 2008 16:48 UTC (Wed) by khim (subscriber, #9252) [Link] As for in-ram-linking-as-modification: just look at the modification licence, its framed in terms of conditions that are really only relevant to fixed, distributable copies. Bingo. And what does it mean? Right: you go back to copyright law. Does it allow this particular modification or not? YMMV. You've been warned. That this has been the prevailing wisdom for a long time (e.g. the GPL FAQ, that NVidia are happy to distribute binary-only drivers for users to load with their GPL kernel, etc..) suggests strongly that your lawyer's interpretation is beyond the pale. GPL FAQ says: if Sun will produce internal ZFS driver for Linux GPL will
not force a disclosure - Sun will be free to use it internally. Good, but
hardly relevant. NVidia does not distribute GPLed code at all - so again,
not a problem. As for "prevailing wisdom"... Who's wisdoms is
this? Not mine - that's for sure. When at least one distribution was closed over GPL troubles,
when Linus says A vendor who
distributes non-GPL modules is _not_ protected by the module interface per
se, and should feel very confident that they can show in a court of law
that the code is not derived, when you can see things like Oh, and at least one major distro
has been served with legal papers due to them shipping closed source kernel
drivers, and more are on the way (I'm not sure if it was Kororaa or
not, but I sure as hell know that Novell wrote Most developers of the kernel community consider non-GPL
kernel modules to be infringing on their copyright. Novell does respect
this position, and will no longer distribute non-GPL kernel modules as part
of future products) it sure looks to me like "prevailing wisdom" says
the following: Basically it means: CDDLed ZFS can not be included in kernel. No way, no how. It'll be infringment - plain and simple. Separate distribution is probably Ok, but watch your back. Oh, and distributors can not include it in distribution too. Do you really think there are a future for ZFS in a position like this? Permanently ostracized without any hope to be ever included in the trunk and without any support from kernel developers? I sincerly doubt it. If your lawers claim that when you are interpreting the license you can safely ignore opinion of guys who actually wrote the code under said license and still keep the copyright... then you are beyond caring, I afraid... Linus and Co always said one thing and thing only: we can not force GPL on non-derived work. No way, not how. And this is the only reason why non-GPLed modules can exist.
Prevailing wisdom? May be in your head... Posted Oct 15, 2008 17:24 UTC (Wed) by paulj (subscriber, #341) [Link]
Firstly, I don't speak for Sun at all, at all... I have never talked to Sun legal about the CDDL - so there aren't even lines to read between. I have no special knowledge, bar a modicum of knowledge of the internal culture.
Your long post is interesting, and I agree that Linux contributors can not incorporate GPL-incompat code because of GPL licence concerns - presuming they can not agree to allow themselves to do so. To be clear: at no stage have I argued that point, and I have no interest in discussing it.
My main point here was to counter the FUD that Sun would sue for CDDL infringement.
Anyway, dinner :)
FOSS and multiple competing commercial interests Posted Oct 14, 2008 16:44 UTC (Tue) by vonbrand (subscriber, #4458) [Link] No, this isn't new. The X Window System started out as a bunch of hardware vendors banding together to develop a common, network-transparent graphics infrastructure.
FOSS and multiple competing commercial interests Posted Oct 14, 2008 17:03 UTC (Tue) by kripkenstein (subscriber, #43281) [Link]
Thanks for the information. I didn't know much about the history of X, so I read the Wikipedia entry now, following your post. Very interesting.
FOSS and multiple competing commercial interests Posted Oct 14, 2008 23:57 UTC (Tue) by masoncl (subscriber, #47138) [Link]
Btrfs has already been released under the same GPLv2 license used by most of the kernel. It has a number of copyright holders outside of Oracle already as well.
Having multiple companies involved definitely helps lower the cost of Btrfs, but that's only a small part of the reason why I tried to get distros and partners on board once I was sure btrfs would work.
Btrfs isn't a specialized FS, it is aiming at a general purpose market with a very broad user base. Pulling in partners and involving them in the design and development allows us to make a better filesystem, and makes it more likely they will ship and support the end result.
FOSS and multiple competing commercial interests Posted Oct 15, 2008 5:09 UTC (Wed) by kripkenstein (subscriber, #43281) [Link]
Thanks for the information. It does make a lot of sense that involving more parties will lead to more adoption.
Can you comment more specifically about the choice of GPLv2 instead of a more permissive license (which might have led to even greater adoption, albeit outside of Linux)?
FOSS and multiple competing commercial interests Posted Oct 15, 2008 13:19 UTC (Wed) by masoncl (subscriber, #47138) [Link]
The Linux filesystems copy quite a bit of code and ideas between themselves. Btrfs includes code from ext2, ext3, reiserfs and xfs, and this wouldn't be possible if I had chosen a different license.
Try Nilfs, now Posted Oct 14, 2008 18:30 UTC (Tue) by hisdad (subscriber, #5375) [Link]
If you are keen on snapshoting and other features,
try www.nilfs.org.
Its quite simple to set up and seems to work fine.
Unlike the others, its been around since 2005.
--John
Try Nilfs, now Posted Oct 15, 2008 0:36 UTC (Wed) by SEJeff (subscriber, #51588) [Link]
And it is slllllllllllllllllooooooooooooooowww
Try Nilfs, now Posted Oct 15, 2008 0:45 UTC (Wed) by hisdad (subscriber, #5375) [Link]
Speed is always relative.
I tried it with postgres 8.3, inserting 1M records into a 60M record table. Time taken went from 528sec (xfs) to 555 sec (nilfs).
Thats slower, not slooooooooooer!
I don't think its well suited to that, but it does a great job on on my cyrus mail spool.
-john
|
|||||||||||||