Submitted by Perhaps overcome with excitement (and forgetting that Apple doesn’t like such pre-emptive disclosures), Sun’s Jonathan Schwartz announced today at Sun event in D.C. that Apple would be making ZFS ‘the file system’ in OSX 10.5 Leopard. “In fact, this week you’ll see that Apple is announcing at their Worldwide Developer Conference that ZFS has become the file system in Mac OS X.”
For the most part I am not surprised, since the alphas have had early implementations of ZFS. But great news to have it confirmed. Having used ZFS extensively in the past I can say that ZFS is unparalleled in the filesystem space. It will be a great addition, especially to their servers. I am keeping my fingers crossed that this will work in Time Machine.
This is actually pretty significant news! But I really didn’t expect it. I thought we’d see ZFS in Leopard+1 or OS XI if such a thing ever exists.
Wow… cool!
Wonder if Apple could give the Mac GUI to Sun Solaris in return for DTrace and ZFS.
That would be even cooler!
Hmm…No.
That is the dumbest thing I’ve ever heard. The work on GUI is a lot more extensive than dtrace and the filesystem.
It sure coule be said the ‘craziest’ idea.
The ‘dumbest’ idea would be if OS X asks for the Solaris GUI
BRB.
No ACL support though?
ZFS has ACL support.
What it doesn’t have is Mandatory Access Control, but then neither does NTFS or HFS+.
<a href=”http://en.wikipedia.org/wiki/Comparison_of_file_systems“>source…
Edited 2007-06-06 21:33
Ummmmmm MAC is not part of the filesystem, it is part of the kernel. Linux uses SELinux for it’s implementation of MAC (Mandatory Access Control) and stores the information in EA (Extended Attributes). However, not all filesystems support EA.
The creator of that table in wikipedia was likely confused. SELinux users can use ext3 as their filesystem because it supports EA.
Solaris uses Trusted Extensions as their form of MAC and I am not qualified to say how it works. If Solaris’s Trusted Extensions need’s extended attributes on the filesystem (which they probably do) then TE should work on ZFS.
Well Linux and other *nixes do support ACLs they are just less talked about than standard Unix permissions on *nix platforms.
I plan to start using them on my Linux system as they offer more flexible file and directory access control options over simple Unix permissions.
Edited 2007-06-07 04:01
Prefered commercial desktop OS: Mac 10.5
Prefered Free desktop OS: Linux
Please no flaming, nothing behind a simple opinion here
Now, I wonder if ZFS will come/is coming to Linux? I have the impression that it was already in some BSD variant… at least in testing.
Just stop asking!
ZFS is not going to be IN the Linux kernel any time soon due to licensing problem.
A FUSE port in is progress, and is actually working well, though.
http://zfs-on-fuse.blogspot.com/
You are referring to FreeBSD.
It is functioning well actually and it will be included in FreeBSD 7.
I do not know about linux there are other options other people has referenced you might want to look for something else.
A FUSE solution would be just as good as a kernel one. FUSE is fast and stable just by judging the stable NTFS-3G driver.
Anyway, I wonder what advantages Solaris has left now that the other *nixes are adopting all it’s showoff features (ZFS, DTrace, …)?
Sun wants to see a Solaris rival against the other *nix platforms, no? Well having all it’s competitive features mirrored in other OSes wont help accomplish that.
Edited 2007-06-07 03:34
If that’s the case, why bother open sourcing anything at all? Someone will come along and copy the features of your product and then you’ll have competitors sporting the same features you have.
Competitors may be able to implement it on their own version of *nix, but Sun came up with the original technology. Ergo, they are the most knowledgable about the tech and will be the ones who will advance it the most. Others will be playing catch-up.
Besides, Sun can always build a business of supporting Solaris for enterprise customers. Sun and the Solaris brand name still carry a lot of weight, even more so in the light of Sun’s positive karma from all the open sourcing it’s doing.
Sun wants to see a Solaris rival against the other *nix platforms, no? Well having all it’s competitive features mirrored in other OSes wont help accomplish that.
Sun doesn’t care too much about OSX, FreeBSD or all the other unices around.
They are afraid of Linux, and that’s why they accurately wrote the ZFS (and other key technologies) license to make it incompatible with the Linux kernel, still being an open license.
Sun is obviously pushing Solaris in the same battlefield where Linux is now, and they don’t way to loose their competitive weapons.
It is sad from a Linux user point of view, but I think that this is a winning strategy for Sun.
Sun wants to see a Solaris rival against the other *nix platforms, no? Well having all it’s competitive features mirrored in other OSes wont help accomplish that.
Sun wants to see Sun hardware and services beat out all others. I can’t imagine they care too much what OS you run, as long as you run it on Sun hardware with a Sun support contract. Otherwise why would they be offering both Linux and Windows as an option on much of their hardware.
Jonathan Schwartz definitely dropped that news to get some recognition in early. Apple will no doubt hype ZFS to the moon and neglect to mention who created it. Sun would love to keep ZFS to themselves, but that’s the way the open source cookie crumbles.
ZFS isn’t a magic bullet, it isn’t mature compared to every other major file system, it still has bug fixes coming in all the time, and it has new features on the way.
Don’t expect to see support for booting from ZFS. It is so early in development that test builds of Solaris don’t even have support for it in the installer.
don´t forget that solaris and osx has different ways of booting and layout of partitions.
osx could very well be able to boot up on zfs.
That comment makes absolutely no sense given that Sun, by their own choice, paid millions of dollars to certify and open the source to Solaris, including ZFS. That doesn’t sound like someone that wants to “keep it ZFS to themselves.”
Yes, and there are still bug fixes to NTFS, and still bug fixes to ext3, and still bug fixes to ReiserFS, and XFS, and so on…your point?
Bzzzt. Wrong. The OpenSolaris version of GRUB already supports ZFS booting. It’s the installer that doesn’t currently. You can boot from ZFS.
Edited 2007-06-06 23:57 UTC
Yeah, that’s what I said
Your logic is incredible.
Sun would love to keep ZFS to themselves, but that’s the way the open source cookie crumbles.
You should run that statement by the many Sun-employeed engineers who cheered the announcement that ZFS is in FreeBSD now, and applaud Apple’s prominent inclusion of it in Leopard.
ZFS isn’t a magic bullet, it isn’t mature compared to every other major file system, it still has bug fixes coming in all the time, and it has new features on the way
If you judge software stability by time and not by practice, you might just be right. But software isn’t like wine, where the older it is the better it gets. In many cases, it can be the opposite. Bug fixes all the time? Show me a popular file system which goes for release upon release without bug fixes. I think this statement of yours illustrates your unfamiliarity with the providence of ZFS. Sun developed and release the initial rev of ZFS with stability as the primary goal, which the 50TB of disk I have under ZFS control would happily agree with. The overwhelming majority of code changes to ZFS since its release have focused on performance and features. Why is it like this? Well, if you wan a new fs which emphisized performance over stability when it left the gate, you are probably one who would enjoy things such as reiserfs *spit*.
Don’t expect to see support for booting from ZFS. It is so early in development that test builds of Solaris don’t even have support for it in the installer.
Alright, now your unfamiliarity with Solaris really shines through with this bit. Tell me how the Solaris installer has anything to do with Apple’s ZFS implementation ?
Well, they apparently want stability based on things like “I’ve used it for six years and it’s never failed me yet”, versus “I’ve used it for six months and it’s never failed me yet”, or “It’s theoretically amazingly sound”.
Now, all of those could apply to the same filesystem, just at different points along its lifespan. But we don’t get to know which ones get ‘time-tested stability’ until they’ve been time-tested…
At least, I think that’s how this goes. I’ve had the same confusion you have about this whole thing, especially when it’s mostly used to justify the existence of the relatively featureless and performanceless filesystems, as if that’s the only way to judge usefulness.
Take it easy my zealous padre. Sun is a lot bigger than some project engineers.
I hope your confidence is well-placed and ZFS turns out to be the messiah of file systems in 5 months when Mac OS 10.5 is released.
> Tell me how the Solaris installer has anything to do with Apple’s ZFS implementation ?
I’d like to believe that the creators of ZFS won’t be beaten to the punch by a third party that is late to the party. But it’s possible!
Who said Sun wanted to keep ZFS to themselves? what Sun don’t want people to do is steal bits of code from Solaris and then stop any enhancements from coming back. That is the problem with ZFS in Linx – the fact that all enhancements would then be licenced under GPL and Sun would be screwed.
Sorry, opensource is more than just running around screaming “GPL! GPL! GPL!” – it actually involves a community where sharing of code occurs in BOTH directions, not just one big giant sucking sound comeing from one corner.
But at the same time, it is mature enough for people to use in everyday situations.
If you have a look at the arc caselog, the new installer is actually ready to be merged into Solaris, which should mean support for bootable ZFS support.
I haven’t seen that. What I do know is that the first version of the installer is due next month: http://opensolaris.org/os/project/caiman/Dwarf/;jsessionid=E8A0FCC9… , and it will be used in the next SXDE, the release date of which I don’t know.
If you spend a little time navigating the site, the caselog sits out like a sore thumb:
http://opensolaris.org/os/community/arc/caselog/2007
Call that the ‘waiting line’ for things to get merged into Solaris.
You have to be joking about that sticking out — I’ve spent hours reading at opensolaris.org and I’ve never even been there. On that community page alone there are hundreds of links to other pages.
I agree that ZFS is not mature compared to “senior” file system but many months ago I’ve read that BeFS guru was working at Apple.
Someone has an idea of what has Apple done with his (I’m speaking of BeFS guru) marvellous file system (perhaps the best BeOS feature) ?
If I am not mistaken, and I usually AM, the “BeFS guy” was the gent that made spotlight happen.
How does this new file system benefit the end user?
@ WorknMan
From what I’ve read, it means no file size limit, same access to a small directory as to the largest directory you can even imagine, practically infinite number of metadata to attach to a file for searching capabilities and organization, no data corruption (recovery and repair on the fly), faster access to the data (time not dependent on the amount of files, data you have on disk), fastest operations for the OS and for user space by using an intelligent cache method instead of traditional virtual memory (begone the pre-binding or pre-fetching tricks), no spinning wheels ever.
Actually, ZFS providing a much more optimized I/O and disk operation the battery life time should actually improve quite significantly.
And this just to look at the immediately visibly stuff.
Edited 2007-06-06 21:41
Could it be true that all features of the ZFS filesystem also have the compromise of needing (much) more RAM and CPU, thus shortening battery life, for more speed and better cache.
It might not be the best for a laptop.
@ Chicobaud
If you read in the end of my comment, I say that ZFS providing a much more optimized I/O and disk operation the battery life time should actually improve quite significantly.
Now that Apple is also fitting LED displays and Intel delivering even more efficient less power-hungry CPUs, I believe generally the battery life can improve a lot.
Remember this is a file system, not a piece of hardware or an application.
Edited 2007-06-07 13:37
ZFS is *very* RAM hungry. Using it on 32-bit systems with 512 MB or less can lead to kernel panics. The sweat spot seems to be around/above 1 GB of RAM.
On 64-bit systems, it’s a lot smoother and runs better, even with just 512 MB RAM.
It can also lead to a lot of disk I/O.
See the archives of the FreeBSD stable and current mailing lists for more info (mostly FreeBSD specific, but with links to Solaris info).
Assuming by end user you mean desktop user there are a few advantages. Among them better data integrity in cases like power outages in the middle of writing to the disk. Fast and effective snapshots of partitions allowing rolling back your disk to a previous state. Also some very nice handling of things like striping a partition across several disks, so that when you add a new disk to your machine you can expand your existing partition onto it.
However to really use all the features of ZFS you really need to be a sysadmin with an Xserver and a few Xraids.
– built to avoid data corruption
– builtin compression
– snapshots (think timemachine without extra disks)
– easily extend fs onto new disks added to system
– software RAID, performance gain when using 2 or more disks
Easy to administer and powerfull, sounds like a match for Apple if you ask me 🙂
I’ve been expecting this since they demo’ed Time Machine.
This alone would qualify for a complete pass on the 4 month delay. 😉
Well done Apple!
Now… announcing this ahead of Steve Jobs… I think Jonathan Schwartz can expect a call from him!
Although given the situation, there isn’t really anything Apple could do beyond making a cranky phonecall to Schwartz.
Steve will probably announce that it’ll just be yet another FS in Leopard, if only to make it clear who calls the shots.
… I’m not that much into the “yay: new filesystem!” anymore. It’s not as if HFS+ currently wouldn’t work. It’s great that there is work being done on the filesystem front, but I’d rather have a tested and true FS on my work-system than something which might seem to bring new trouble etc.
Don’t forget that Apple introduced true case-sensitivity as an HFS+ option in Tiger. But if you actually went for it, you suddenly noticed that major apps (plus quite a few little ones) had trouble with it. Among them, you suddenly couldn’t launch Adobe CS apps anymore. So implementing a new FS might have hidden traps for end users.
Steve will probably announce that it’ll just be yet another FS in Leopard
Steve? Man, you Apple fanboys are such whores.
That’s his name. What did you expect, that we’d call him “Wonka”?
Maybe he felt you were referring to him with too much familiarity… next time try “Steven Paul Jobs, Esq.”
🙂
That’s the best comment i have every read, i still can’t stop laughing.
It isn’t that HFS doesn’t not work, it’s that it doesn’t not suck
Don’t you mean, “It isn’t that HFS+ doesn’t work, it’s that it doesn’t not suck.”
Oh, damnit, I’m not even sure when I don’t know what I don’t mean anymore….
…head…exploding…
Edited 2007-06-07 15:12
To tell the truth, I haven’t not the slightest idea no more.
This isn’t even being tested yet? I would think something as important/fundamental as the underlying file system would be seeded to testers well in advance of a final release. Doesn’t a new FS need to be tested more intensively than say, a backup program like Time Machine?
Edited 2007-06-06 21:49
of course it’s been in testing. apple has been testing it for some time. just like they made dual build of OSX on intel boxes. they plan ahead and explore lots of options. and i am absolutly sure that time machine and ZFS are interconnected. if you go back far enough and look at the times the rumors popped up about both of them they are right on the same time line (there are of course other reasons i think this but theres just one for thought).
Well of course I’d assume this has been tested in-house, but the usefulness of that only goes so far, which is why they have beta testing programs. I was thinking it was odd to not have widely tested, outside their lab conditions, something like a new FS yet.
“Time Machine” is probably just a GUI for ZFS snapshots.
Windows and Linux have similar snapshot features of their own, but with GUIs more sketchy than what Apple will probably provide.
The snapshot concept is an unintuitive one at first, before you grasp what is going on. Basically the OS constantly monitors the file system, and only updates snapshots retroactively as their view on the world is changed. So if I take a snapshot of my file system every 5 minutes but change no files all day, then no extra space (other than negligible overhead) will be taken up.
When I delete or change a file after taking those snapshots, one copy of that file is kept, which is then pointed to by every snapshot that was taken when the file was in that state.
So on a computer with plenty of CPU, memory, hard drive space, and non-epic amounts of file system changes (ie most PCs), automatic snapshots can be taken all the time. Then when you want to retrieve a backup, there could be a GUI that shows you snapshots and file versions, as well as being able to right-click on a file and see the different versions of it that are available.
I would expect since the cat is out of the bag, and it will likely be announced at the end of June, developers can expect a shiny new build with lots of surprises.
Beta cu could have had some agreement to sign, so they can’t talk about it even if they wanted.
This is great news for Mac OS X IMO!
Congrats, Mac fans. You’ve come a long way in the last few years.
I’m a Linux/Ext[3|4] fan, myself.
But ZFS definitely has an appeal.
ZFS is a bit beyond the pale for Linux for a number of reasons. Licensing being one. But the biggie is that ZFS’s attitude WRT the layering of functionality are totally alien to the Linux kernel’s philosophy.
But no matter. You and Sun are definitely the ones to beat, now. (That’s a compliment, BTW.)
Expect some rhetoric followed by some code from the Linux camp. 😉
We have good solid filesystems. And I have no problem with the idea of us playing catch up for a bit.
I sincerely believe that most of the *real* progress made in this industry is related to someone chasing someone else’s taillights.
Thanks for providing a couple of nice ones. 😉
Edited 2007-06-06 22:12
I can’t mod you up any further
seems one of the “top secret” features is clear now – and it’s amazing!
Should be Schwartz instead of Schwarz
May the Schwartz be with you!
😉
As far as I know, Because apple is using EFI it could just write a driver so that the Firmware could read ZFS. There would be no need to wait for bootable version of ZFS because of the flexibility of EFI.
Also, it was commented on the Ars Technica that there is a project has fast-tracked case insensitive support in ZFS. It seems likely we could see ZFS as a primary FS in Mac os 10.5. I can’t wait.
Luminair, As far as I know ZFS does not need to take snapshots. Since writes changes to the file in new space, and does not write over the old file. Cloning in ZFS could make apps like iPhoto and Aperture save large amount of space as they share the base data and only annotate changes.
It’s possible that boot support will be in at the start, but I don’t see it being the default on Apple systems right away. It’s a big change, going from 0 to all-zfs at one time.
ZFS definitely does snapshots, the command is “zfs snapshot”
http://docs.sun.com/app/docs/doc/819-5461/6n7ht6qsa?a=view
… HABEMUS FILESYSTEM!!!
OH YEAH!!! 😀
-5 ? Well that’ll teach you to make latin/catholic jokes I guess.
For those who didn’t get it, it translates to “I announce to you a great joy – we have a filesystem”. From the traditional latin phrase to announce a new pope. ( http://en.wikipedia.org/wiki/Habemus_Papam )
No really off-topic, although the caps are a bit much.
“No really off-topic, although the caps are a bit much.”
Not much. The papal announce is screamed.
Caps is scream. 🙂
P.S.
I’m Italian and I’m studing Religious Sciences, a roman catholic pastoral theology five-years degree. ;-P
After reading up on ZFS one of its big advantages is to easily add more disk space to the pool. Will we see Apple announcing some type of network storage device in the future?
An interesting architecture change got fast tracked at opensolaris recently :
http://www.opensolaris.org/os/community/arc/caselog/2007/244/;jsess…
ZFS case-insensitive support, Submitted 27 April 2007
“Any given file system volume (i.e. mount-point) may support both case-insensitive or case-sensitive behavior. The caller must be able to specify the behavior on lookup and create operations.”
Ostensibly for CIFS support, but helps when having it as the default fs for OSX (traditionally not case sensitive) too.
Edit: damn “Forgotten” beat me to it above. Didn’t see it, oh well.
Edited 2007-06-07 05:28 UTC
I guess it’s what the Macs need to speed them up.
Ican’t say I’ve been impressed with the performance of OSX.
Needs lots of RAM and still not very snappy even on hardware that’s considered high-end these days (dual core systems).
As for Sun, they can play their games with licenses that are incompatible with the Linux kernel.
It will not help them fool people.
In the real world people are talking about Linux, nobody even knows what Solaris is, and nobody cares.
And I see Sun signing an agreement with Microsoft soon, like Novell and Xandros did recently.
It’s an industry wide conspiracy against Linux.
They can’t make money from Linux so they’ll be trying to kill it anyway they can.
But in the end freedom that only Linux provides will win.
I guess it’s what the Macs need to speed them up.
?
Ican’t say I’ve been impressed with the performance of OSX.
Neither am I. Still, I dare you to provide me with a similarly equipped operating system that can run as comfortably on my 450Mhz G4 Cube.
Needs lots of RAM and still not very snappy even on hardware that’s considered high-end these days (dual core systems).
See above.
As for Sun, they can play their games with licenses that are incompatible with the Linux kernel.
It will not help them fool people.
In the real world people are talking about Linux, nobody even knows what Solaris is, and nobody cares.
And still, many of the Solaris tech is finding its way into other operating systems, and Linux is left out of the blue, forced to re-invent the wheel.
It’s an industry wide conspiracy against Linux.
Black helicopters! They are everywhere!
Edited 2007-06-07 10:42
I’m not too familiar with Linux distros that run on G4 but I think Debian might.
To make a comparison though, let’s take a typical Pentium 3 with 256Mb RAM system (quite old now) and both Slackware and Debian with KDE as a desktop will fly beautifully on it.
I don’t knon what you mean by “comfortably”, speed or features of the OS, features of the applications, or both?
But I can say this.
Although I use OSX only ocasionally I can tell that KDE as a desktop has many times the functionality of OSX.
If we also take into account the apps, then yeah, we lose some polish and features (arguably) but not that much and it’s a small trade-off that we gain back in freedom.
What Solaris tech you mean specifically? Dtrace, ZFS?
These are only usefull to very few people.
Besides Linux has had good alternatives for a long time whereas Solaris is just inventing these recently to compete against Linux.
For over a decade Sun let Solaris fossilize like a dynosaur and now that nobody is buying them they are suddenly having a garage sale.
Well, too bad! It’s too little and too late. And it’s not free enough. If it’s not GPLed it’s not free. Simle as that.
Well, SCO was a Linux distributor and then suddenly did it’s best to hurt Linux.
Novell is “sleeping with the enemy” and now Xandros joined them too.
Will Sun resist when Microsoft shoves a mountain of money in front of them?
So yeah, black helicopters are everywhere when billions are at stake.
Although I use OSX only ocasionally I can tell that KDE as a desktop has many times the functionality of OSX.
Indeed, you really use OSX occasionally for saying that …
I’ll bite. How much experience do you have with Solaris, or are you another one of those Linux users who make comments about Solaris without any experience to back them up?
Just because Solaris might not have a “killer desktop” doesn’t mean it isn’t feature rich. I administer Solaris and RedHat systems on a daily basis and if it was my call, I would show RedHat the door.
Pawel Jakub Dawidek did the porting for FreeBSD. A 100% port! Because of the FreeBSD base in MacOS, it’s rather easy for them to “port” what’s left
The coming MAC (mandatory access control) in Leopard is from BSD too and in FreeBSD available since FBSD 5. Apple will of course hype it, without mentioning where they got it from.
Since many people here are talking about ZFS exclusion to Linux, I have a question about the CDDL. From the CDDL FAQ:
Yes, you can modify and redistribute code licensed under the CDDL, including charging for it if you wish. However, you will need to meet the terms of the CDDL, including making the source of your modified code available under the CDDL.
So no matter what Apple does to the code, Sun can implement their (Apple’s) changes because the source will be available to the public (Sun)? And Linux can’t use the code (ZFS) because it must be licensed under CDDL? I don’t want to start a license war, but am only looking for some answers. Also, there were rumors that OpenSolaris would go GPL v3, which would mean CDDL. If so, wouldn’t Apple have a hard time using ZFS in future products? Would they have to create a “fork” of ZFS, using/modifying the last version released under the CDDL before Sun switched to GPL v3?
Edited 2007-06-07 14:19
Also, there were rumors that OpenSolaris would go GPL v3, which would mean CDDL. If so, wouldn’t Apple have a hard time using ZFS in future products?
I was thinking the exact same thing. But there must be a way around it, otherwise Schwartz wouldn’t be so gung-ho about the GPL3, right? Anyone else care to enlighten or conjecture?
Apple is obligated to release the source for any modifications they make to the zfs code, but Sun likely wouldn’t incorporate it. They require copyright assignment from contributors; the code can remain in the public but Sun won’t formally integrate it into Solaris without that.
The incompatibilities with linux boil down to the CDDL having features/restrictions (depending on your POV) that are incompatible with the GPL. There are also a number of patents involved with zfs right now, and at the moment Sun’s patent provisions apply only to CDDL licensed code.
If Sun releases openSolaris as GPLv3, it will remain dual-licensed with CDDL. So Apple/FreeBSD will still be free to use the code, they’ll simply do so under the terms of the CDDL without needing to meet GPL provisions.
> So no matter what Apple does to the code, Sun can
> implement their (Apple’s) changes because the
> source will be available to the public (Sun)?
It seems that way, but the CDDL license doesn’t mandate that distributing binaries compiled from CDDL licensed code (like Apple are) give the recipient (Mac OS X users) the right to view that code. This means Apple needn’t make their modifications public in the first place, just that *if they do* it must be under the CDDL. At least that is my interpretation of a quick skim of the license.
> And Linux can’t use the code (ZFS) because it must
> be licensed under CDDL?
Correct, but theres nothing (except sanity) stopping someone from creating a GPL’d implementation of ZFS if they can do so from the file system specification without looking at CDDL licensed code.
> there were rumors that OpenSolaris would go GPL v3,
> which would mean CDDL. If so, wouldn’t Apple have a
> hard time using ZFS in future products? Would
> they have to create a “fork” of ZFS,
Not necessarily. For example, all official GNU projects have contributors sign over copyright to the Free Software Foundation, giving the FSF/GNU the option to relicense code anytime they please, even if the old and new licenses are incompatible.
Presumably Sun hold all copyright to any development done internally on their own ZFS implementation, and so they wouldn’t need to ‘fork’ as such, just cease releasing ZFS under the CDDL and re-tag the code base with the GPLv3 license.
* This means Apple still has the choice to negotiate a private license agreement with Sun without involving the wider open source community at all *
** It seems that way, but the CDDL license doesn’t mandate that distributing binaries compiled from CDDL licensed code (like Apple are) give the recipient (Mac OS X users) the right to view that code. This means Apple needn’t make their modifications public in the first place, just that *if they do* it must be under the CDDL. At least that is my interpretation of a quick skim of the license. **
The CDDL is a per-file license, and with the CDDL you’re only bound on the actual program files, so if you incorporate CDDL files as part of a cumulative application, you’re only obligated to release code modifications on the original CDDL files. This is in contrast to the GPL copyleft, which extends GPL licensing to any and all additional program files linking to the original files. But I think the real incompatibilities are due to requirements such as attribution notices etc.
From that perspective, the CDDL is far closer to the LGPL than to the GPL. The intent is that if a party improves upon the original cDDL code, they’re obligated to share, but they’re not necessarily obligated to share applications built around the CDDL code.
** And Linux can’t use the code (ZFS) because it must
be licensed under CDDL?
Correct, but theres nothing (except sanity) stopping someone from creating a GPL’d implementation of ZFS if they can do so from the file system specification without looking at CDDL licensed code. **
Well, at the moment zfs is heavily patented and “licensed” use of those patents only extends to the original CDDL implementation. While Linus advises developers to generally ignore patents, it’s not likely that they would accept code that is so clearly encumbered, regardless of the community’s viewpoint on patents. Sun would have to extend their patent provisions to OSS in general, the way IBM has, rather than restricting them to CDDL.
Alternatively, there’s probably nothing stopping third-parties from packaging zfs in such a way that it can be easily compiled against the linux kernel; the GPL would restrict distribution of precompiled packages linked against the kernel and would prevent the inclusion of such code in mainline, but GPL doesn’t restrict what users actually do, and neither would the CDDL. It would simply be an extra step, similar to the ATI/nVidia driver dance many users do.
edit: futile attempt to fix !$*#$% quote tags. I gave up.
Edited 2007-06-07 16:37 UTC
> edit: futile attempt to fix !$*#$% quote tags. I gave up.
^ Using good old email reply syntax works!
Edit: Wtf, HTML identities seem to work properly in previews but when you actually post they are shown as is :/
Edited 2007-06-07 22:09
There’s no guarantee that Apple is distributing ZFS under the terms of the CDDL. It’s true that ZFS is licensed under the CDDL. It’s also true that Sun own the copyrights to all the ZFS code so Sun can license ZFS however they want. It’s entirely possible that Apple licensed ZFS under totally different terms.
See you in Hell HFS+ (Journaled) !
It’s already supported in Opensolaris, the main issue from what I know was that the Solaris installer didn’t support it.
Nextenta have just released a “Core Platform” release that has a ZFS/boot enabled installer.
http://www.opensolaris.org/jive/thread.jspa?threadID=32407&tstart=0
I discovered this problem few days ago about ZFS Fragmentation.
There is some discussions on this topic on opensolaris.org
http://www.opensolaris.org/jive/search.jspa?objID=f80&q=fragmentati…
In short, Copy On Write is a good feature on the paper, but when you use it in production on high I/O server, it puts your server on it’s knees. In fact, I am surprised that this design ‘flaw’ was never discussed before.
Just check how it works from the Sun presentation.
Everything is copy-on-write: “Never overwrite live data”
So when you modify a file with ZFS, it will never erase some part of it. Instead it will write the new datas on free space and then modify the file blocks to use the new datas and keep the old one for the ‘constant-time snapshot’.
But what happens here ? ZFS just fragment the file !!! Just imagine that there is no free space for gigabytes before you can write something. Your file was contiguous and with a simple modification on the file some parts of it will be moved at the end of the hard drive. And this happening each time you modify a file. Just imagine the nightmare one year later on the hard drive.
Some people report that just reading files use 50% of the CPU time of it’s server, not to say that ZFS disable some optimisations you will do with oracle. From one of the thread above.
The purpose of zero-filled tablespace is to prevent fragmentation by
future writes, in the case when multiple tablespaces are being
updated/filled on the same disk, correct?
This becomes pointless on ZFS, since it never overwrites the same
pre-allocated block, i.e. the tablespace becomes fragmented in that
case no matter what.
And the most worrying thing is the answer given by the community leader of ZFS just few months ago
We have some ideas, one of which is to periodically do an on-line
“defrag” by relocating most of the blocks. This is obviously not the
most elegant solution, but it may work. We’re currently investigating…
–mattw
http://www.opensolaris.org/jive/thread.jspa?messageID=93437𖳽
We were about to use it in production at my work, I will warn them about that and make some real world bench before deciding something.
If you read this thread, you find out that the problem is with the NFS client, not ZFS:
http://www.opensolaris.org/jive/thread.jspa?threadID=24014
I have Solaris 10 1/06 and 6/06 machines in production running ZFS volumes and have yet to have any problems with them.
Yes, and on this one, the ZFS community leader acknowledge the problem.
http://www.opensolaris.org/jive/thread.jspa?messageID=93437𖳽
Considering the discussion is several months old and no new comments have been made since indicates one of the following:
1. That the fragmentation issue only affects a limited number of users with specific configurations.
2. That it has been corrected by patches (see the README for the Solaris 10 Recommended Patch Cluster for more details).
3. That interest in the issue fell off for one reason or another.
I think it is a combination of (1) and (2).
Is this going to stop me from using it on either Solaris or MacOS X, no. In the case of Solaris the alternative that provides part of the functionality of ZFS is Veritas Foundation Suite which has a tiered pricing structure based on the architecture of the machine you are going to run it on. For example, to run VxVM on a V240 would cost over $6,000.00 for the license.
Considering the alternatives for MacOS X, I would still use ZFS because I feel the benefits of ZFS outweigh the detractors.
To correct my earlier post, my Solaris 10 1/06 machine does not use ZFS since it was released with Solaris 10 6/06.
Either way, I’m sure Apple has done their research when it comes to ZFS if they intend to feature it in any meaningful way.
Exactly. I am sure Apple weighed a number of options before they decided on ZFS.
I hope. Perhaps they even modify the code
I found a good answer to this problem from Bill Moore the co leader of the ZFS Team.
http://www.opensolaris.org/jive/thread.jspa?messageID=14997
Quite interesting, the problem seems marginal.
For a “marginal problem” it didn’t seem to stop you from spreading the usual “there’s a serious problem with ZFS” FUD. And this series of posts are written months before the link you provided was written in the first place.
If I was reading about hundreds or thousands of users complaining about fragmentation with ZFS I would take it seriously, I see a handful of posts and I think unique circumstances. Not everybody is going to get the same mileage out of ZFS based on what they are doing and what equipment they have. And to highlight the problems a few people are having with ZFS and make it out to be far more serious than it actually is is simply wrong!
If I was making FUD:
1) I would never post this thread from Bill Moore
2) I wouldn’t try to debate about the problem
3) I would just say that ZFS suck without trying to find an answer
After that, if you considere each opinion that’s not yours is a FUD, so yeah, it’s a FUD …
Actually all you have done is try to support your position until you found another post that contradicted it and posted it before someone else did, I don’t call that debating.
At least my opinion is based on actual daily use of ZFS in production as opposed to people who post that have no relevant experience with ZFS. I think I would rather listen to someone who has experience with a particular product than someone who can just point out what they think is wrong with it.
Yeah, because you think that I will trust an unknown guy on a forum claiming he had an experience of daily use of ZFS in production ?
Could you explain to me how I am sure that you are not lying ?
I can imagine that:
“hey boss just look at this post on osnews, this guy is running ZFS on production servers, just go for it without checking anything …”
Be careful what you wish for:
http://www.osnews.com/story.php/15911/A-Look-at-Solaris-10-606
You want the ZFS information on my production systems as well or is this sufficient?
Well, your article learned me nothing I didn’t already known from the Sun’s communications/articles. Your article is just like an entry in ablog. Nothing highly technical.
Bill Moore = a Sun engineers
Robert Escue = a guy who wrote an article on OSnews.
Excuse me if I need to heard a Sun engineer instead of you.
Actually, I have written 3 articles for OSNews on Solaris 10, I’m sure you can find them. You can attribute the lack of detail (if you say so) to the limited time given to produce the article. I had Sun backing on each of them, although I turned down their offers for some assistance.
For the last four years I have also been an External Beta Tester for Sun. And while I might not be an engineer, I feel that my background is a little higher than some of the people who post here who have noting more than a one time install experience.
Now, what are your qualifications since you seem so insistent in questioning mine.
So now if I were to get a Mac and triple boot Windows OSX and Linux…
Linux will only be able to read the Windows NTFS partition.
Linux won’t be able to do anything with the OSX partition (unless I use FUSE).
Windows won’t have access to anything other than itself.
And I’m not sure about OSX, I’ve never used it.
Is that correct?
So now if I were to get a Mac and triple boot Windows OSX and Linux…
Linux will only be able to read the Windows NTFS partition.
Linux won’t be able to do anything with the OSX partition (unless I use FUSE).
Windows won’t have access to anything other than itself.
And I’m not sure about OSX, I’ve never used it.
Is that correct?
not entirely.
Linux can safely write to NTFS thanks to the NTFS-3G fuse module, altough I believe it has some issues with compressed /encrypted files. Beside that, it is believed to be stable enough to be used.
And, i believe that there’s something for windows that enables you to read (and maybe write) ext3 fs (that is the default on several distros).
…I really don’t mind taking fanboy heat from anyone as long as Jobs continues to (even if it isn’t innovation) wrap bigger and better things into his operating system.
I don’t care if solaris had X first. I don’t care if MS had X, Y or X first. I don’t care if XXXXX started with xerox parc labs… whatever. Mac OS X has them – and it makes my 8-12 hours a day spent on my computer a better experience.
Now I eagerly await some fantastic new hardware announcements to go along with OS X and its improvements.