ZFS Hits an Important Milestone, Version 0.6.1 Released

sfcrazy writes "ZFS on Linux has reached what Brian Behlendorf calls an important milestone with the official 0.6.1 release. Version 0.6.1 not only brings the usual bug fixes but also introduces a new property called 'snapdev.' Brian explains, 'The snapdev property was introduced to control the visibility of zvol snapshot devices and may be set to either visible or hidden. When set to hidden, which is the default, zvol snapshot devices will not be created under /dev/. To gain access to these devices the property must be set to visible. This behavior is analogous to the existing snapdir property.'"

Comments

Attaching a non-rc version number...

[Anonymous Coward]

does not a milestone make. Looking at this issue list - https://github.com/zfsonlinux/zfs/issues - makes me wary to even consider zfs on linux for any serious work.

Kernel panics, deadlocks, data corruption; not really things you'd want.

Re:Attaching a non-rc version number...

[Anonymous Coward]

I've had 6 production ZFS servers under heavy load for over a year (FreeBSD 9.0-RC -> 9.1-RELEASE) without any problems. I've started building all of my new servers with root-on-zfs to start taking advantage of beadm (boot environments, lets you do a clone of your root file system, do an upgrade in a jail, then try booting off it, and then decide if you want to keep it, or roll back)

Re:Attaching a non-rc version number...

[Anonymous Coward]

Keep in mind that zfsonlinux is different than ZFS in *BSD, and any testaments to the stability of ZFS in *BSD are impertinent.

Re:Attaching a non-rc version number...

[Anonymous Psychopath]

Keep in mind that zfsonlinux is different than ZFS in *BSD, and any testaments to the stability of ZFS in *BSD are impertinent.

Or irrelevant. But true enough.

Re:Attaching a non-rc version number...

[Anonymous Coward]

If you haven't had some data loss with zfs testing, you haven't tested enough. Might I suggest a set of nice rare earth magnets?

http://www.magnet4less.com/index.php?cPath=1&gclid=CNGdmPCzorYCFYI-MgodL3YA5Q

Re:Attaching a non-rc version number...

[dbIII]

I started giving it a go last year on an old 32 bit server with 24 IDE disk bays and just threw whatever random IDE drives that happened to be lying around into it. It copes very well with hardware failures, and if you group the disks by age/size/model it performs well even if some of the disks are very slow and old. After I threw out all the dead and dying disks the thing was actually useful as a second holding area for data while reinstalling and reformatting a production system. "zfs send" is an awesome way to move a lot of data around, and even that old box could move a lot back onto the reformatted system. Of course it chews power like nothing else so is turned off, but it was a good thing to play with to work out how to implement freebsd and zfs on the production servers.

Re:Attaching a non-rc version number...

[ulzeraj]

Whoa! They've ported beadm to FreeBSD too? I feel stupid for not knowing it.

That tool is awesome. You can actually chose what version of the operating system you want to boot (from ZFS clones and snapshots) from the boot loader.

Re:ZFS need Linux maturity

[Lennie]

And so does btrfs.

Re:So... I presume this is a file system.

[Anonymous Psychopath]

That's the kind of information that could be mentioned in the summary.

Isn't that kind of like saying articles about the sun should mention it's a star?

Re:ZFS need Linux maturity

[kthreadd]

ZFS runs great on FreeBSD as well.

Re:So... I presume this is a file system.

[Anonymous Coward]

Seriously? Is this how far we've fallen? Return your geek card at the door and never return.

I often agree with complaints about little-known acronyms, products, or projects not being explained in Slashdot summaries; but this? FFS!

Not ZFS 0.6.1...

[Alcoholic Synonymous]

Sorry, but the title is misleading. ZFS did not hit 0.6.1, only this port for Linux. ZFS uses it's own versioning, which actually recently bumped to v5000.

Re:Not ZFS 0.6.1...

[gtirloni]

Let me know when it's over 9000 though.

Re:So... I presume this is a file system.

[Arancaytar]

And this "FFS", this is also a file system? ( :P )

Re:So... I presume this is a file system.

[Guy Harris]

And this "FFS", this is also a file system? ( :P )

Yes.

Re:So... I presume this is a file system.

[Guy Harris]

What's TFS? Another file system?

Yes.

Depends on which ZFS

[SIGBUS]

Version 5000 is used for community ZFS implementations that have feature flags (Illumos, BSD, and Linux).

If you're talking about Solaris, the current version is 34; any version past 28 comes after Oracle closed off Solaris. Note that beyond version 28, the community and Oracle ZFS pools are not interoperable.

Re:When did Slashdot becom Digg?

[chromas]

Yesterday. We do complain about the stories not belonging here—now, it's our fault because we didn't vote hard enough. Or maybe they're integrating the firehose.

Legal Issues?

[halfdan the black]

Wasn't there some legal issue where you could not use ZFS with Linux, where you are breaking the law by doing so? Couldn't Oracle or FSF, or someone else sue you for using ZFS on Linux?

Re:Legal Issues?

[armanox]

No. The issue is the license(CDDL) and GPL don't play together in a way that lets you ship CDDL software with GPL software (but the other way around is fine).

Re:Legal Issues?

[Anonymous Coward]

Sort of, but it's different than you are thinking. Linux is licensed under the GPL 2, while ZFS is licensed under CDDL. Those two licenses are not compatible. Since the GPL is a distribution license, not a use license, there is nothing stopping you from using ZFS on Linux. However, you can't ship the two combined as you would then be violating the license. The practical effect is that you won't ever see a kernel implementation of ZFS ship with a Distro unless oracle relicenses ZFS. You'll have to download, compile, and install ZFS yourself for the Linux-based computers that you want to use it on. And that's perfectly legal within the scope of the licenses.

Re:Legal Issues?

[Anonymous Coward]

Sort of, but it's different than you are thinking. Linux is licensed under the GPL 2, while ZFS is licensed under CDDL. Those two licenses are not compatible. Since the GPL is a distribution license, not a use license, there is nothing stopping you from using ZFS on Linux. However, you can't ship the two combined as you would then be violating the license. The practical effect is that you won't ever see a kernel implementation of ZFS ship with a Distro unless oracle relicenses ZFS. You'll have to download, compile, and install ZFS yourself for the Linux-based computers that you want to use it on. And that's perfectly legal within the scope of the licenses.

That is only an issue with monolithic kernels. Sabayon Linux and Gentoo Linux are currently shipping ZFS binary modules on their live media.

Re:Legal Issues?

[Guspaz]

It's also worth noting that the primary platform for ZFSonLinux is Ubuntu, where DKMS is used to dynamically compile the kernel module. I believe the same is true for most other distros that aren't source-based.

Translation: you add the repository and install the "zfs" package and it does everything for you. So installing ZFS is no more difficult than installing any other package, and the licensing issues are completely irrelevant to users.

Re:Legal Issues?

[cas2000]

The practical effect is that you won't ever see a kernel implementation of ZFS ship with a Distro unless oracle relicenses ZFS. You'll have to download, compile, and install ZFS yourself for the Linux-based computers that you want to use it on. And that's perfectly legal within the scope of the licenses.

A zfs-dkms module package is also possible, which automates the compile and install of the kernel module.

There are un-official packages for ubuntu and debian already which do this (spl-dkms and zfs-dkms for the kernel modules, plus some other packages for libs and utilities) - this is how i run zfs-on-linux on my debian systems: I git pull the source repo for dajhorn's ubuntu packages and rebuild them for debian. Only one minor edit is needed to pkg-zfs/debian/control, change the dependancy on 'zfs-grub' to 'zfs-grub | grub'....since I don't boot from ZFS this works well for me. I think it would work any because the grub package in debian has zfs support built in. Once they're built I can install them with 'dpkg -i'....they install with as little trouble as any other package (they just take longer because they have to be compiled against the current kernel-header package). if the dkms packages were apt-gettable, it would be even easier.

There's no legal impediment to such packages being officially in debian (which already has similar dkms packages for other free-but-not-GPL-compatible modules, and debian even has some xxxx-dkms packages in non-free for nvidia and fglrx and similar non-free kernel modules).

As long as the kernel and ZFS are distributed not as a combined work but separately which the end-user has to combine themselves, then there is no derivative being distributed (the GPL's goal is not to prevent users from exercising their freedom to use software as they please, its goal is to prevent GPL code being distributed embedded in proprietary software)

BTW, debian also already has a zfs-fuse package which contains the CDDL zfs code plus dual-licensed GPL2 and CDDL packaging scripts. Debian also has a Debian GNU/kfreeBSD port which can have zfs natively without any licensing issues to work around (CDDL is compatible with BSD license, but not with GPL)

Other distros tend to care less about the niceties of adhering to licenses than debian does so may take some short-cuts - but my bet is that they'll do something very similar to the above to avoid giving Oracle any excuse for suing them.

Re:Legal Issues?

[cas2000]

That is only an issue with monolithic kernels.

I wouldn't want to bet on that. Compiling and linking CDDL module source against the kernel's GPL source almost certainly creates a derived work (it's complicated but, in short: depends on whether it just uses header information or actually links in kernel code).

Distributing a CDDL source-code plus scripts package (e.g. a zfs-dkms package) which compiles the zfs kernel modules on the user's own machine as a result of the user's explicit action (i.e. to install zfs-dkms package) still creates a derivative work but because the user is NOT re-distributing that derivative, it does not violate the terms of the GPL.

The GPL only restricts distribution, it does not restrict usage...this is not a bug in the GPL, it is a design feature...it's the way it's supposed to work.

Sabayon Linux and Gentoo Linux are currently shipping ZFS binary modules on their live media.

Good luck to them - that's not a legal risk i'd want to take. Any one of the kernel's many contributors could be offended enough to sue.

The greater risk is Oracle seeing an excuse to sue - they're a huge corporate with a highly-competitive no-holds-barred attitude, lots of money and attack-lawyers on staff, and they're not only the owners of the ZFS copyright, they're also contributors to the kernel.

Huge corporates care less about bad publicity and PR backlash than most people think, they care only as much as it has the ability to have a sustained long-term effect on their bottom-line....and even then they only care if they've got a mass-market product that's vulnerable to mass boycott, or their activities bring on the ire of one of the large, mainstream activist groups.

The risk of annoying geeks with a low-profile (no children killed or baby animals tortured or rivers being polluted or particular group of people being discriminated against) issue like software licensing really won't stop them. We're not Oracle's target market, anyway - at best we're useful-but-optional marketing auxilliaries to provide some "geek cred" for their product....their target market is other corporations.

Re:Legal Issues?

[cas2000]

While I am not a lawyer, I have actually done reading on this topic and I am certain that almost everything you wrote is wrong

that's nice for you. I'm certain you don't have a clue what you're talking about.

you certainly don't know enough to be encouraging anyone to bet against an expensive lawsuit - which doesn't actually have to win in order to succeed, it just has to bankrupt you (in which case, it wins by default because you don't afford lawyers to defend yourself any more)

Re:Legal Issues?

[cas2000]

what is it about software licensing that brings out the nutcases?

fuck off back to under your rock, troll.

your loony conspiracy theories aren't needed, aren't helping and certainly aren't wanted.

Re:File systems have version#?

[armanox]

I suppose that FAT12, FAT16, FAT32, exFAT, ext2, ext3, ext4, reiserfs and reiserfs4 all confuse you already?

Actually, I think it's just a version on the driver.

Such important milestone and version number 0.6.1?

[enriquevagu]

Why not labeling it 1.0? Looks like it is still in beta...

is that sarcasm? or lack of knowledge?

[fireylord]

THe answer is yes, yes it is :p
http://en.wikipedia.org/wiki/Amiga_Fast_File_System

Re:File systems have version#?

[Anonymous Coward]

NTFS is also separately versioned on Windows, so this is not that uncommon.

Why ZFS?

[Gothmolly]

What are the advantages of ZFS over, say, ext4? If you have a low-memory machine and not a lot of storage, does it buy you anything?

Re:Why ZFS?

[Anonymous Coward]

Snapshots, volumes, checksums, easy expansion, better drive management, shorter recovery time when the system crashes, etc etc etc. ZFS is so far ahead of ext4 they are in completely different fields. ZFS works well on low resource machines too. I use it on a small home server with 512MB of RAM and it's been running great for over a year.

Re:Why ZFS?

[Gothmolly]

I get that - but why? Why is it worth the additional hassle?

Re:Why ZFS?

[icebraining]

You can get much of that if you add LVM to ext4, though.

Re:Why ZFS?

[Guspaz]

Better data integrity? Checksums on all blocks means the OS can tell if data is corrupt, and the data can be seamlessly recovered from redundancy (typically parity from raidz or raidz2, which also doesn't have the raid5 write hole because ZFS is copy-on-write).

Easier to use? zfs management happens through the "zfs" and "zpool" commands which are generally much easier to work with than obscure necromancy commands required for traditional types of systems that make me care about cylinders and partitions.

More flexible? The storage pool method, where you build a pool of capacity and allocate filesystems out of it, gives you a great deal of flexibility and simplicity. I just keep adding more storage to my pool as required, either by adding more RAID arrays or increasing the size of disks in those arrays, and then I've got my primary filesystem for storage, I've got a deduplicated one I use for backups, and I've got a compressed one I use for long-term archives. And creating a new one like that takes about five seconds without having to repartition or reformat anything. Creating/deleting filesystems is about as much effort as creating/deleting files.

Easier snapshots? Snapshots are instant on copy-on-write filesystems. Any modification of data causes the block to be copied anyhow, so all a snapshot has to do is not delete older blocks.

ZFS is one of a handful of next-gen filesystems (along with BTRFS and HAMMER) that are so far beyond traditional filesystems that it's a really eye-opening experience using them. That's not to say ZFS is perfect, or that the ZFSonLinux implementation is perfect, but it's in a reasonable state of stability at this point, and the advantages that these new filesystems offer is substantial.

I do wish that ZFS had asynchronous deduplication like HAMMER, though. ZFS deduplication requires atrocious amounts of RAM (estimates go from 5 to 20 gigabytes of RAM per terabyte of deduplicated data), while HAMMER has effectively no extra memory required at runtime for dedupe, because it just scans the disk afterwards and does the deduplication after the fact, so it doesn't need to hold the full block table in memory at all times.

Re:Why ZFS?

[dbIII]

Huge - but mostly it's about the difference between LVM and ZFS in a multi-disk situation.
Snapshotting and portability is the biggest difference I can think of with ext4, everything else is pretty well comparing a LVM+ext4 bundle with zfs.

If you have a low-memory machine and not a lot of storage, does it buy you anything

I'd say not, ZFS is not something I'd run on a machine with less than 2GB.

Re:Why ZFS?

[dbIII]

It's less hassle with a muti-disk setup than other forms of RAID but more hassle with one disk, especially on linux where you have to download the kernel modules separately.

Re:Why ZFS?

[smash]

Lvm + ext4 does not do end to end checksums. KVM + ext4 does not make every write full-stripe.

Re:Why ZFS?

[smash]

That's LVM of course. Typo... A major difference is that the file system is aware of the underlying backing store and can use that to its advantage, due to the "layering violation" stuff that some Linux developers decided was inherently bad and unacceptable.

Re:Why ZFS?

[smash]

Async dedup is a trade off I guess. It means you need a window to run your dedup in where performance will no doubt suffer. If you are doing it inline you pay the memory or SSD cost to hold the dedup table, but don't need to worry about scheduling a maintenance window.

Re:Why ZFS?

[smash]

If you're running it o. Pc-bsd it isn't any extra hassle.

Re:Why ZFS?

[TheRaven64]

I can't speak for Linux, but on FreeBSD it's no extra hassle. First, you don't need to think about partitioning, you just create a big zpool. You can restrict the size of filesystems within it, but they're not hard limits, so it's very easy to expand them. On the other hand, you do get the benefits of having different partitions (i.e. you can optimise them for different use cases, e.g. turning on compression and deduplication on a volume that you don't access much and turning them off in filesystems that require more speed).

Some things are just nice to have in the background, like the end-to-end checksums, so you can detect single-block corruption, and RAID-Z being aware of the structure of the filesystem (unlike RAID-5), so recovery time is proportional to the amount of data, not to the size of the disks. It can also easily take advantage of SSDs for acceleration. If you add SSDs as either cache or log devices then it will speed up reads or burst writes significantly, even if most of the pool is on spinning disks.

The biggest win for me is O(1) snapshots. You can snapshot a volume trivially and it has no performance impact (well, that's not quite true: some things become slower if you have tens of thousands of snapshots) and you can trivially roll back. The entire filesystem is copy-on-write. You can also clone a filesystem if you want to experiment. This is very useful for jails or VMs, where you can have a single filesystem with an up-to-date base install and then just clone it (constant time, takes under a second) to create a new clean VM.

Re:Why ZFS?

[TheRaven64]

I'd say not, ZFS is not something I'd run on a machine with less than 2GB.

It depends a lot on the size of the disk. The general rule of thumb for good performance with ZFS is 1GB of RAM for each TB of storage (more if you're doing dedup). I stuck ZFS (PC-BSD) on an old laptop I gave to my tango group to play music. It only had 1GB of RAM, but it only had 20GB of disk space, so it was completely fine, and it means that if they unplug it (the battery is dead) by accident without shutting down then they still have a consistent filesystem, and I could snapshot it in a known-good state that they could refer to if they broke everything.

Re:9000

[Anomalyst]

Stop, Dave...I can feel it.

zfs hits an important milestone

[Anomalyst]

that's gonna leave a mark

ZFS Linux perhaps

[tyrione]

But sure as hell doesn't mean ZFS like the twit who titled this thread as reaching an important 0.61 milestone. I'll pass on Linux with ZFS. I'll use FreeBSD where it is mature.

Re:ZFS Linux perhaps

[Guspaz]

Consider that all three major ZFS platforms (Linux/FreeBSD/Illumos) are working on a common core that they all share, and that the lions share of ZFS development is coming from the Linux community. Perhaps the Linux ZFS community should not be dismissed so readily.

FreeBSD is a fantastic platform for ZFS, but considering that both FreeBSD and ZoL are pulling down new work all the time, it's not automatically more stable.

Re:ZFS need Linux maturity

[dbIII]

I agree. I tried ZFS on linux with disappointing results (got hammered by half a dozen users at once and slowed to a crawl) and I'm geting vastly better performance on the same hardware with freeBSD for now (takes the same load and more without slowing to a crawl - even while doing a scrub at the same time), but I'm sure it will catch up with a bit of work. I've still got one linux box with ZFS which gets a job done (offsite redundant storage) - it's slow in comparison but does appear to be stable.

The very nice thing is that when it is I'll be able to do a "zfs send" to just dump entire filesystems onto whatever system understands zfs. The only drawback of having it on solaris or freebsd is that people that have only been exposed to linux will take a little while to get up to speed on a slightly different OS if I get hit by a bus.
I love these file systems that check themselves while running and so don't take a fsck of a long time to come up after a power outage. A UPS doesn't save you from workplace health and safety mandated electrical system checks :(

Re:So... I presume this is a file system.

[smash]

2 of them in fact, Berkeley and Amiga - not the same.

Re:ZFS need Linux maturity

[utkonos]

Is it production ready yet? Or is it still in a beta/testing state?

Re:File systems have version#?

[unixisc]

Should we then go back to what the file system was for the punch cards?

Re:ZFS need Linux maturity

[CAIMLAS]

It does, with a few caveats: namely, you need to have one of about 3 disk controllers to make it stable. That's the biggest one. And preferably, run it from something which has been specifically designed for ZFS, like FreeNAS instead of as 'freebsd' itself, since that's just a moving target...

Re:ZFS need Linux maturity

[CAIMLAS]

How long ago did you try it? I had similar problems very briefly when I tried it at around 0. 5.8 I believe. Performance is much better now.

Personally, I've got 3 ZoL systems here at home: a Phenom II x3 with 16GB of RAM (which is my main VM host), an AMD Bobcat with 8GB (storage/backup mainly), and an AMD XP 3200+ with 8GB of RAM (secondary/failover VM host) - all run virtualization extensions, and of the half dozen VMs which are getting used regularly (and another 8 which are always on serving various network functions - logging, SMTP, monitoring, DNS/dhcp, etc.). I've got an SSD as ZIL, and that may be part of the reason why it performs as well as it does - all of the systems are lightning fast and responsive.

Re:ZFS need Linux maturity

[dbIII]

How long ago did you try it?

In December with whatever was current then I had the system that just could not cope under linux with ZFS and then I reinstalled it as freebsd. I've got a another linux ZFS system that is running OK but it never gets to work hard. It fine so long as there are not a lot of things going on, but with only half a dozen users trying to read things off the same filesystem at once it slowed down to almost nothing. Maybe it would have been OK with more than 4GB (however the current one has 16GB and has similar problems though not as bad) but it just could not cope with the linux version of ZFS that was available in December.

Re:So... I presume this is a file system.

[TheRaven64]

but this? FFS

No, this ZFS. FFS is much older and does a lot less.