I am surprised that I am the first person who has noticed this little fact...

Both Client and Server versions of the OS X build that was released at WWDC lack ZFS support.

No longer are you able to create, view, mount or do much of anything with ZFS in Snow Leopard.

Given that Apple explicitly removed the ZFS read support that existed in Leopard, I find it *highly* unlikely that Apple will ever find themselves returning to this highly advanced file system.

Bottom line is that file systems really aren't all that exciting on their own, they do enable for some very cool things to be done in the software applications sitting above. What ZFS gave you was block level snapshots... an unlimited number of them in a space efficient, maintainable way... and the ability to do software RAID that in many cases outperformed hardware RAID.

There are plenty of other benefits, however the vast majority of them are duplicated under any "new" file system.

It is very unfortunate that Oracle bought Sun recently, as I would greatly expect that under Oracle's hands, ZFS will either die to their BTRFS that they are developing or be merged together into a new mashup of ZFS with a slightly lower memory footprint for the re-silvering process and a GPL license to go along with it.

It's a sad day, but it really is what it is.

~Adrian

Is there any specific reason why Apple removed the ZFS? And what exactly, have they replaced it with?
I hope ZFS ends up under a GNU license, with the fancy silver removed.

ADS has been 'replaced' with HFS+ with compression (or rather, that's what they're saying as it wasn't in the WWDC Build.

Now that Oracle owns ZFS, while GPL is possible, it seems like it's more likely that BTRFS will pick up a few new on-staff developers, and then we can see something that is usable in another 2-3 years.

Not like Apple would allow GPLed code 'into' their kernel...
Much like Linux won't let CDDL play.

~Adrian

I'm sorry, but I'm a newbie on these things. How exactly is this new (Brtfs) approach better than the old ZFS?
What are the possible performance gains as such?

Bottom line is there are no benefits of BTRFS at this point for non-ext3/4 users. BTRFS is essentially an extension to EXT4 that allows for snapshotti g and block level mirrors.

There is also talk of incorporating volume management as is seen in ZFS.

The key benefit is that BTRFS can be released under the GPL license, thereby allowing it's inclusion in the mainline tree of the Linux kernel.

ZFS was licensed by Sun under the CDDL, which allowed for it's inclusion in the Solaris and BSD variants (including OS X) but explicitly kept Linux out of the game.

In a few years, BTRFS *may* become a viable system, however at the moment it is really just ext4 with with some ZFS functionality that people were able to copy when all they really bothered to find out was what Sun put in their press releases.

Okay, that was a little harsh... I am sure that the BTRFS developers are very competent, but their product is still years from completion, and then there is ZFS which has been stable for the past 5 years.

~Adrian

didn't even notice that Oracle bought Sun, what a bummer.
Thanks for sharing your opinion and knowledge, very interesting.

I'm a Solaris sysadmin and I do like ZFS a lot. I personally use OpenSolaris in my workstation because of that.

I think one of the reasons of that decision would be that ZFS is not so suitable for a desktop machine. IMHO (in my humble opinion), of course.

That is because the read performance under some workloads is questionable and to help reduce that impact exists ARC (Adaptive Replacement Cache), whom uses all of the available RAM for its purposes.
To a user with a 2GB computer (and not technically proficient) having just a few MBs free could be a little bit shocking.
In the real world ARC will adapt that memory consumption as needed but it's not hassle free.

Two of the strong points on using ZFS: ease of administration and major reliability, are not that appealing to end users, and at the end of the day most of the ZFS features doesn't represent real improvements to end users (at least not as appealing as performance improvements).

How many of us have personal computers with mirrored disks, lots of RAM or a spare disk to configure as a zpool cache vdev?

I think it's a good decision by the managers in short term for the desktop.
And I also think it's a must have on the server side.

It'd be real nice to have it as an optional native filesystem on the OS in future releases.

Well, given the fact that Adobe's applications will still refuse to run under HFS+ Case Sensitive, it is hard to believe that Apple would elect to add another optional file system to their client OS.

Time Machine is a very strong case for ZFS on the Client, but as Apple has fully pulled even read support from the client OS, it seems as though they have simply given up on the technology.

Disappointed, yes... Gonna loose sleep over it, no.

I guess ZFS will stay on my file servers only for the foreseeable future.

~Adrian

ZFS has been removed from the GUI. The ZFS filesystem controls and kext's are still in the OS and you can still format partitions ZFS, you'll just have to do it from the command line.

Cheers

( kextstat | grep zfs )

Other than being 100% inaccurate, your logic is impeccable bjr2.

At no point did Snow Leopard have a working graphical administration tool.
Following WWDC, Apple removed all ZFS management commands (zfs & zpool), and then in the latest build proceeded to remove /System/Library/PrivateFrameworks/ZFSManager.framework.

Nice try, though.