Submitted by Another turn in the Apple-ZFS saga. Apple has made available a developer preview of ZFS for Mac OS X with read/write capability. The preview is available to all ADC members. From the readme file: “ZFS is a new filesystem from Sun Microsystems which has been ported by Apple to Mac OS X. The initial (10.5.0) release of Leopard will restrict ZFS to read-only, so no ZFS pools or filesystems can be modified or created. This Developer Preview will enable full read/write capability, which includes the creation/destruction of ZFS pools and filesystems.” Update: Will it ever end? The release has been pulled from ADC by Apple.
Cool! Seems like ZFS is getting adopted by most systems now. I’m happy to see my two favorite systems offering it: OS X and FreeBSD.
That was more than predictable. Apple had ZFS on their agenda for Leopard, but they couldn’t deliver early enough for the developer preview.
Now they are too late to make it the default FS, but it’s just reasonable to continue development and introduce it a little bit late.
I’m guessing in one of the update packages (maybe 10.5.1 even) ZFS read/write will make it to the masses. Mac fanboys will hail this as Mac upgrades being worth the price due to the amazing features included in the incremental updates. Right now I’m a little torn if I will be upgrading, I really want their virtual desktop implementation, but without ZFS I don’t see any other killer features for me.
I think for me it is all the little incremental improvements that I will switch for.
AFAIK ZFS is not bootable. Apple can’t make it the default FS until then.
ZFS is bootable, but it is a new feature that is largely untested and unimplemented in OS installers. ZFS is still pretty immature on Solaris, so you can see why it will take Apple some time to port it and roll it out for primetime use.
Edited 2007-06-26 14:25 UTC
That was more than predictable. Apple had ZFS on their agenda for Leopard, but they couldn’t deliver early enough for the developer preview.
ZFS was never on the agenda for Leopard at all. The rationale behind ZFS for Apple was large storage arrays for servers and their Xserves. It will be a long time, if ever, that we see it on a desktop.
ZFS is still new, it still isn’t a bootable filesystem yet, not to mention the inevitable quirks and incompatibilities that come with dropping something as critical in as a new filesystem.
It’s a pity MS have done their old tricks of reinventing existing tech otherwise we would have ourselves near enough a platform indipendent FS
Then, from your POV, it’s a pity Sun doesn’t relicenses under a suitable license. In order to become a “platform independent FS”, ZFS would need to be released under the BSD or something. With the CDDL, it’s just no possible.
Of course they’re not doing that, because they’d lose the patent-related stuff of the CDDL (welcome to real bussiness)
And then, there’s people who thinks that ZFS is a great filesystem, but not the best possible. Sun’s marketing has worked quite well to make people think that ZFS is somehow a sort of blessed filesystem, when the reality is that it’s just another. And not the last one.
Edited 2007-06-26 14:07
I think you’ve taken my comments some what out of hand – At no point did i even suggest ZFS should go GPL (or whatever license you assumed I implied).
I simply said it’s a pity ZFS has not become platform independent (ie you can plug a HDD in an system and it will read it – at the moment, only FAT32 offers that).
It’s also a pity some OSNews reads jump to the offensive when composing responces rather than thinking about the actual post
Interesting, the only people complaining about ZFS licencing is the Linux crew; FreeBSD, MacOS X and a few other operating systems don’t seem to have a problem with it.
Maybe the Linux ‘crew’ should rexamin the licence in which it is licenced under instead of blid loyalty to the GPL.
Interesting, the only people complaining about ZFS licencing is the Linux crew; FreeBSD, MacOS X and a few other operating systems don’t seem to have a problem with it.
They’ve no problem using it, however they’ve lots of problems in depending on it and making it the ‘primary’ filesystem, because this would break the possibility of taking a copy of freebsd and relicensing it or use it in your propietary product – you couldn’t do that with the CDDL as you do it with the BSD.
Don’t fool yourself, CDDL is more GPL-ish than BSD-ish. And this issue is _really_ important for the BSD community, who always aimed at having a completely BSDish operative system. For them, having such important piece of the system as the filesystem is under a non-BSD license _is_ a problem (which is why they’ve rewritten GNU tools like ls and cp under a BSD license!).
Edited 2007-06-26 15:04
NO, it does not. The CDDL code is only used for building a ZFS kernel module. It doesn’t taint the FreeBSD kernel in any way.
BTW, I can’t see how that would pose a problem to Linux even, since Linus himself used the AFS example so countless many times to state his “benevolence” towards certain binary kernel modules. How would that be different with ZFS? The layering violation stuff they keep repeating which Jeff Bonwick, the guy who lead ZFS development, already addressed quite a number of times?
When layering makes sense, it does. When it doesn’t, it doesn’t. Sticking to a layering model for the sake of “purity” is stupid. The TCP/IP stack model bypassed half of the OSI layering and did what actually made more sense. I really didn’t see anyone (but the OSI guys) shed a tear. Guess which model is easier to program to.
Edit: “when” -> “which”, 2nd parag. =P
Edited 2007-06-26 18:38
BTW, I can’t see how that would pose a problem to Linux even, since Linus himself used the AFS example so countless many times to state his “benevolence” towards certain binary kernel modules. How would that be different with ZFS?
Linus would be fine with a binary ZFS module, but you don’t actually expect him to code and maintain it, do you?
When layering makes sense, it does. When it doesn’t, it doesn’t. Sticking to a layering model for the sake of “purity” is stupid.
I tend to agree with you here, but maintainability can be an issue for a project like Linux that has so many different file systems. If each of them did their own thing instead of sharing code it becomes a lot more difficult to change any related part of the kernel without introducing bugs.
Why would I? Hasn’t the FUSE one been written already?
What I don’t know is this: FUSE can be compiled into the kernel; this much I know. Or just put the KM on initrd and load it at boot; anyway, what I don’t know is if one can put a FUSE module on initrd and have it loaded at boot time. Is it possible?
Because if it is, this whole quagmire of ZFS and Linux and GPL and CDDL and whatnot (I’m being specific to the ZFS situation, mind you) is mute.
If it’s not, I’m pretty sure this should be doable and someone is probably working on this already. One can run BusyBox, TFTP servers, hotplug agents and whatnot from initrd, so why not a FUSE module?
Hasn’t the FUSE one been written already?
Yes. Right now it only runs at about half speed, but the NTFS driver shows that it can be improved. I’m not sure if you could actually boot off of a FUSE fs or not, but that’s kind of a moot point right now anyway since ZFS still doesn’t support it. It seems like it could at least be possible though.
ZFS does support booting. GRUB must be patched, though. Sun’s patches to GRUB are available from the OpenSolaris site.
ZFS does support booting. GRUB must be patched, though. Sun’s patches to GRUB are available from the OpenSolaris site.
My mistake. Solaris still doesn’t support it, though, right? I’m not ready to call it production-ready until that happens.
ON (OpenSolaris/Nevada) supports it as of either b61 or b63. What’s still missing is support for it in the installer, but that’s afaik in testing and waiting to be merged.
>which is why they’ve rewritten GNU tools like ls and cp under a BSD license!
Nonsene! There was an ls and cp before any GNU tools. (BSD since 1978). So you don’t have to write anything new what’s already there. OpenBSD does this with some original GNU tools.
>For them, having such important piece of the system as the filesystem is under a non-BSD license _is_ a problem
In *BSD community people don’t have as many problems as someone may think.
Interesting, the only people complaining about ZFS licencing is the Linux crew; FreeBSD, MacOS X and a few other operating systems don’t seem to have a problem with it.
They don’t have a problem because they all use licenses that mean that CDDL licensed stuff can be imported and used.
Maybe the Linux ‘crew’ should rexamin the licence in which it is licenced under instead of blid loyalty to the GPL.
You can give us all a call when any of the BSDs, or Apple, have an open source kernel of the size, quality, and sheer number of developers and support that Linux has. To be honest, as interesting as ZFS might be, I doubt whether anyone involved with Linux would want to sacrifice those fundamentally good things, or the license that helped create them, in order to have a port of ZFS.
Additionally, given that ZFS has quite a few layering violations, that covers LVM and RAID functionality, perhaps even worse than Reiser4, it would also need some picking apart to be imported into the Linux kernel.
You can give us all a call when any of the BSDs, or Apple, have an open source kernel of the size, quality, and sheer number of developers and support that Linux has.
Only zealotery here
To be honest
From a zealot, I highly doubt that …
Additionally, given that ZFS has quite a few layering violations, that covers LVM and RAID functionality, perhaps even worse than Reiser4, it would also need some picking apart to be imported into the Linux kernel.
Yes,that’s why Sun said ZFS eliminates all the old storage concepts.
Anyway, why should we follow the LVM and RAID paradigm forever ? God wrote it ?
You seems like an old procedural programmer criticizing the apparition of OOP…
Only zealotery here
Jealousy or uncomfortable squirming in the seat doesn’t advance the discussion I’m afraid. The fact is that the GPL has played a big part in individual developers, companies (a lot of them competitors) and organisations contributing to the Linux codebase and feeling like they’re doing it with everyone’s cards on the table on a level playing field.
That’s what Linus wanted, that’s why he chose the GPL, quite apart from any FSF ideologies and politics – and you’d have to say that it has worked, given the platforms Linux has been ported to and how widely it is now used in comparison to any other open source kernel.
Like I said, I doubt whether Linus or any of the other kernel developers would sacrifice that just to get an implementation of ZFS.
From a zealot, I highly doubt that …
Again, that doesn’t advance any discussion, and that sounds an awful lot like zealotry to me.
Yes,that’s why Sun said ZFS eliminates all the old storage concepts.
Errr, no it doesn’t. It just replicates the functionality LVM and RAID provide in a whole lot of ways.
Anyway, why should we follow the LVM and RAID paradigm forever ?
Because they’re logically different layers in the system? A filesystem is there to specifically store files, and LVM and RAID are effectively containers for those filesystems.
God wrote it ?
What was that about zealotry again?
You seems like an old procedural programmer criticizing the apparition of OOP…
Sorry, but you don’t know what you’re talking about.
It’s interesting how the guy talking about zealotry and without any withstanding arguments gets more votes than you do when you actually give a more to the point answer (and more technically inclined).
I see that ZFS is the near-future FS, but, as you said, LVM+RAID is a perfectly fine solution and competitor (at least as containers).
OK, this meme stopped being funny (but I modded you up anyway because of your points on zealotry). LVM+RAID doesn’t provide nearly the same level of administrative comfort as ZFS. Think: quota management, compression management, snapshot management.
Not to mention features that LVM+RAID simply don’t have.
http://unixconsult.org/zfs_vs_lvm.html for a comparative overview. BTW, if you don’t agree, James is open for corrections.
Edit: “has stopped” -> “stopped”
Edited 2007-06-26 19:40
I see that ZFS is the near-future FS, but, as you said, LVM+RAID is a perfectly fine solution and competitor (at least as containers).
OK, this meme stopped being funny (but I modded you up anyway because of your points on zealotry). LVM+RAID doesn’t provide nearly the same level of administrative comfort as ZFS. Think: quota management, compression management, snapshot management.
I could not agree more. Where I work, we have around 200 TB online. I have used a lot of storage management software and on usability and plain acceptability, I would rank the packages I know something like this:
ZFS & its volume management
HP-UX LVM
Veritas VM (complicated, but flexible)
Windows 2003 VM (simple, great boot disk protection)
Sun SVM (nothing special, but it works just fine)
Linux MD + LVM2
MD + LVM2 is a little awkward but I can live with that; If I couldn’t I would need to find a career other than computing. What is unique about it is the amount of freak-out behavior I have seen with it: failure to boot a degraded mirror on a stock distro setup, failure to destroy snapshots, retention of snapshots past reboots, etc. Some of this was probably my fault, but so far I have found the docs to be awful (try reading a Veritas administrator’s guide before you whine). In the case of the snapshot / reboot problem, docs imply it is impossible. Go figure. Anyway, MD + LVM is the only VM solution I seriously worry about on production systems. At least ext3 is stable.
So if ZFS is layered wrong, I don’t want to be layered right
And if we could find a new glib sound bite to replace “layering violation”, that would be nice too.
It’s interesting how the guy talking about zealotry and without any withstanding arguments gets more votes than you do when you actually give a more to the point answer (and more technically inclined).
Or perhaps it’s because there is a lot of Linux zealots here …
How can you give some arguments to:
“linux is just r0x0r and rulez all OSS OSes because it’s bigger, better and more reliable”.
There is no argument in there dude. Trying to argument with a zealot is pointless whatever OS he’s using.
Edited 2007-06-27 07:17
A more valid question may be, “Why should we break the LVM and RAID paradigm?”. Does ZFS really provide anything which can’t be accomplished just as well via the existing framework? Please note that is a different question than does ZFS currently provide functionality not implement via the LVM and RAID paradigm.
I know that Sun engineers think the answer is yes it does; however the Linux kernel developers seem just as convinced that it doesn’t. Really the only way to know for sure is to let the Linux developers have a try at it.
I think it’s the same answer to the question:
“Why should we create a new Open Source OS (linux) while there is already a lot of them available (BSD) ?”.
The answer:
Because people want to try new things, they want to try new way to approach the same problem.
The two questions are not the same. One question regards starting a new project or fork. The other is trying to force a change on an existing project.
Hey could you not paint us all with the same brush. I really couldn’t care what ZFS is licenced under that’s sun business and its up to them what they do with it.
There’s nothing to reexamine, and the blind loyalty is actually built right into the license. Sun is free to license ZFS however they like. But as it stands there is nothing that the Linux community can do, nothing that we can reexamine, and nothing that we can be more open-minded about that can allow a Linux port of ZFS to be distributed.
I appreciate the CDDL and other MPL-based licenses as they encourage merging and commercial collaboration while also protecting the four freedoms to an extent. Weak-copyleft, share-alike is a place on the free software licensing spectrum that resonates with a developer’s sense of freedom and meritocracy. But this is beside the point.
Sun knows what it needs to do in order to bring ZFS to Linux, and it may be a non-starter. But there’s nothing the Linux community can do to bring ZFS to Linux. No matter what we think of the CDDL, or even the GPLv3, it just isn’t possible. We have the option of either working within the limitations of the GPL or deciding to drop over 6 MLOC and to start over to the tune of about 10,000 man-years. Is this even worth examining?
Considering the prohibitive difficulties involved in changing Linux to be GPLv3 (even should Torvalds et al want to do so), I don’t really see HOW you’d expect them to move to something CDDL compatible… and just so they could use ZFS?
…Unless you’re suggesting they simply ignore the terms of the GPL, which is foolish. You can argue people have blind loyalty to the GPL as some Shining Standard of software licenses, but the not-easily-mutable fact is that Linux is licensed under the GPLv2 and no later, and must be dealt with as such.
For what it’s worth, there was an article on OSNews just a few days ago wherein Linus Torvalds stated that Sun releasing ZFS under the GPLv3 might make relicensing the Linux kernel more tempting.
Still, you know…
One file system for every OS… (FAT32 notwithstanding) would be really nice…
Yes, this is can be downloaded by any ADC member…
Unfortunately it’s only of use to folks with the 10.5 beta installed – it won’t install on 10.4.
Yes, this is can be downloaded by any ADC member…
Unfortunately it’s only of use to folks with the 10.5 beta installed – it won’t install on 10.4.
Well, Leopard has leaked already so I think it will be downloaded quite a bit more than there are official leopard cd’s out there.
In regards to ZFS and Linux.
Check this link.
http://www.linuxworld.com/news/2007/061807-zfs-on-linux.html
Somehow I doubt, the Apple folks did the port themselves. I bet the farm they used most of the code from FreeBsd to complete the project, and they seem to be taking credit that they did it themselves…
I find your post humorous because Jordan Hubbard works at Apple now.
http://lists.apple.com/archives/publicsource-announce/2001/Jun/msg0…
And I find yours even more humorous because Jordan Hubbard if you have not noticed joined in 2001! ZFS was published way after that!
Also, the FreeBSD port was released to CURRENT 2 months back so you do the math. Whats the relation between the link you posted (Jordan joining in 2001), (Mac OS X ZFS port), and (FreeBSD 7 ZFS port)?
I think your reply to “Credit to FreeBsd?” is totally inaccurate. Please, reference something that is related to ZFS and Jordan.
Nope, the timing doesn’t match. There were reports of Apple finding bugs related to how Spotlight traverses directories ahd how that affects ZFS, and that was *way* before Pavel feature-completed the ZFS port to FreeBSD.
The fact that the ZFS port to FreeBSD makes extensive use of GEOM facilities doesn’t help the porting to Mac OS X at all.
Anyway, I’m positive that once Leopard is out you will be able to find Apple’s modifications to the ZFS sources, as those are part of Darwin (and the matching Leopard version, 9.0.x IIRC, will be released) and hence will be available for anyone to see.
So, please, speculating that Apple is not giving credit where it’s due has no basis in reality whatsoever. Bash Apple anyway you like if they fail to give due credit when you can prove they did not.
Edit: “BSD” -> “FreeBSD”, 2nd parag. No other BSD implements the GEOM stack.
Edited 2007-06-26 18:42
This whole thing reminds me of Microsoft and WinFS. I guess Apple and Microsoft aren’t so dissimilar after all, regardless of what Steve Jobs wants everyone to believe.
What the hell are you talking about?
At least Apple is -shipping- their FS work.
Jesus. Where’s WinFS? oh wait… that’s right.