ZFS For Linux Finally Lands In Debian GNU/Linux Repos (softpedia.com) 150
prisoninmate quotes a report from Softpedia: It took the Debian developers many years to finally be able to ship a working version of ZFS for Linux on Debian GNU/Linux. For those not in the known, ZFS on Linux is the official OpenZFS implementation for Linux, which promises to offer native ZFS filesystem support for any Linux kernel-based operating system, currently supporting Arch Linux, Ubuntu, Fedora, Gentoo, Red Hat Enterprise Linux, CentOS, openSUSE, and now Debian. And it looks like their ZFS for Linux implementation borrows a lot of patches from Ubuntu, at least according to the changelog for zfs-linux 0.6.5.6-2, the version that is now available in the unstable channel for Debian users to install and test.
Bunking with Hans? (Score:5, Funny)
I notice that this article was submitted by prisoninmate, yet the filesystem in discussion is ZFS, not ReiserFS.
What gives?
Re: (Score:1)
Re:I'll probably hold out a while longer. (Score:5, Insightful)
I'll use it with FreeBSD.
Re: (Score:2)
For varying definitions of "supported". Hasn't development of the old ZFS been at a standstill apart from critical patches, while OpenZFS has continued to evolve?
Re: (Score:2)
OpenZFS was forked at version 28. Oracle has put out 5 versions since then, the most notable feature being native encryption. I wouldn't call it a standstill, but I wouldn't call it blistering speed either
https://docs.oracle.com/cd/E23... [oracle.com]
Re: (Score:2)
The ZFS encryption was basically finished before Oracle acquired Sun.
Re: (Score:2)
Why do you call it GNU/Linux and not GNU/ZFS/Linux? The filesystem is a pretty darn important part of an operating system.
It is. That's why Linux supports about 30 of them out of the box.
Re: (Score:2)
A lot of them are useless,but four are not:
ext4 may not be the most existing file system, but it is the default, solid, and good enough in most situations.
XFS is faster (especially for big files), but it has a few issues. reiserfs is quite good, and more space efficient for small files. Neither are clearly superior to ext4, unless you have specific requirements.
ZFS is taking it to the next level, and as such it may well be worthwhile. It remains to be seen how reliable it is on Linux, but in terms of featur
Re: I'll probably hold out a while longer. (Score:1)
Re: (Score:1, Flamebait)
Re: (Score:2)
Re: I'll probably hold out a while longer. (Score:1)
Am I the only person who uses BTRFS?
Don't Bother with ZFS (Score:2, Interesting)
If you want the benefits of ZFS use FreeBSD. It's a much nicer/cleaner OS than Debian or the other 3000 linux distros in use cases that require ZFS.
Re: (Score:2)
I hear that all the time, but why would I switch operating system only for the file system? Ever heard of someone switching to Linux to get ext4 support? Or to Windows to get NTFS support? Of course not. File system is a very minor component of an operating system these days. Except if I were building a NAS, I wouldn't even consider making a comparison of file systems when choosing an OS.
Re: (Score:1)
I hear that all the time, but why would I switch operating system only for the file system? Ever heard of someone switching to Linux to get ext4 support? Or to Windows to get NTFS support? Of course not. File system is a very minor component of an operating system these days. Except if I were building a NAS, I wouldn't even consider making a comparison of file systems when choosing an OS.
Go look up containers and jails? ZFS is more than just a filesystem as it can be used for situations with jails. That is a big difference.
Also without SystemD and with old fashioned init you get a much more reliable and production ready system
Re: Don't Bother with ZFS (Score:2)
If you had actually used systemd, you'd know it makes your services more reliable, not less.
Re: (Score:2)
Re: (Score:2)
Have you heard of this new docker thing? Virtualization is so outdated, you make extra copies of the OS?
Re:Don't Bother with ZFS (Score:5, Informative)
If you are building a server, then file system absolutely matters.
Ext4 is fine but limited and not nearly as mature or stable as XFS. When building a server that *matters*.
ZFS blows the doors off of anything the Open Source community has *ever* built (in terms of file systems). It has features that Linux users have been desperate for. For example, file system snapshots and rollbacks. Yes, you can do it with LVM2. Guess what? LVM2 sucks at it (in comparison to ZFS).
Re: (Score:2)
But wait, during the decade or so that ZFS was "almost ready," wasn't there a functional transactional file system option with a less-well-known name?
I'm trying to remember what it was - and also flush out a believer who can tell us: "X" was doing this 5-6 years ago and has been actively improving since then, ZFS is just a better known name, not a better system.
Anybody remember that one?
Re: (Score:2)
For example, file system snapshots and rollbacks
NetBSD does this in a filesystem-agnostic way with its Filesystem snapshot device [gw.com]
Why build the functionality into one particular filesystem? Oh yes, poor design.
Re: Don't Bother with ZFS (Score:1)
Re: (Score:2)
Oh yeah sorry, I forgot. Most servers out there do not use ZFS. And they still perform stuff that matters.
Re: (Score:2)
Re: (Score:2)
Btrfs is not on on par with ZFS. Not even close.
Re: (Score:1)
I hear that all the time, but why would I switch operating system only for the file system? Ever heard of someone switching to Linux to get ext4 support? Or to Windows to get NTFS support? Of course not. File system is a very minor component of an operating system these days. Except if I were building a NAS, I wouldn't even consider making a comparison of file systems when choosing an OS.
For NAS, its a killer RAID filesystem. It's a bit of a memory hog, but the anti-corruption and snapshotting features are great. The on the fly compression doesn't save much space for media files, but that's handy to have also.
My only gripe is how tricky it is to add more drives to the raid. I don't know if I would switch operating systems to have it, but I'm really liking it and I wouldn't want to go back to another filesystem.
Re: (Score:2)
It would be interesting for the ability to add more drives dynamically, perhaps with a new RAID type that can work with dynamic adding and removing of drives, minimizing the time the array is in a degraded state.
My "ideal" of a filesystem that sits on BSD is EMC Isilon's OneFS or NetApp's WAFL filesystem. This not just offers what ZFS has, but the ability to link other nodes via Infiniband and use their filesystems, presenting a filesystem which is redundant across drives, and redundant across nodes. If a
Re: Don't Bother with ZFS (Score:1)
Re: (Score:2)
--ZFS is much more flexible with mirrored pairs. Build a 5- or 6-disk RAIDZ2, and you have to add the same number of drives ( with the same capacity, e.g. 6x1TB + 6x1TB ) to double the RAID capacity properly and keep your I/O throughput sane.
--However, if you start with a mere 2-disk mirror, you can keep adding mirrored pairs much more easily. Example: Start with 2x1TB WD RED NAS drives in a mirrored pair. Create a mirrored zpool with both disks. You start with ~1TB of redundant mirrored storage, with ~
Re:Don't Bother with ZFS (Score:4, Informative)
Because ZFS is actually part of the kernel structure of FreeBSD. Linux puts the filesystem in userspace which makes ZFS in userspace inefficient. (Hell all filesystems in user space are inefficient, but ZFS doubly so because of what ZFS is.)
ZFS, in short is a bullet-proof file system, if setup correctly. Unfortunately the VAST majority of use cases are overkill. A proper ZFS setup would require no less than 4 mechanical drives and 8GB of RAM. Just for the file system. Because underneath ZFS is essentially a very smart software RAID 6 and journaling system. So you can pretty much kill one drive in a 4 drive system and it will still function as if it was still there, no "degraded mode" like in a hardware raid. Then when you replace the drive, or upgrade it to a larger drive it just works seamlessly, no week-long rebuilds with impaired performance.
Compare that to ext2/ext3/ext4/UFS/exFAT/NTFS which are nothing but filesystems, some with journaling.
Re: (Score:3)
> ZFS, in short is a bullet-proof file system, if setup correctly. ... Because underneath ZFS is essentially a very smart software RAID 6 and journaling system.
Well, considering hardware RAID has a silent corruption bug, I'd say it more then "essentially" :-) Especially with Raid-Z2 and Raid-Z3.
See Pages 13 .. 18
ZFS The Last Word in FileSystems [illumos.org]
Measurements at CERN
* Wrote a simple application to write/verify 1GB file
* Write 1MB, sleep 1 second, etc. until 1GB has been written
* Read 1MB, verify, sleep 1 s
Re: (Score:2)
Uhm, no.
ZFS-on-Linux is the project that creates a kernel module for all the ZFS bits, and integrates the filesystem into the Linux storage stack.
It's not as stable and reliable as the FreeBSD setup, but it's a geek of a lot more stable and performant than the old FUSE-based setup on Linux .
Re: (Score:1)
He's not BS'ing. ZFS on linux has been tried as a userspace filesystem (via FUSE, among other efforts...), and its nowhere near as good as a native, in-kernel implementation. Doing the FS in userspace actually costs more in raw performance.
Re: (Score:2)
The obvious reason would be for a file server. Other situations are not so obvious, but as for minor - unlikely.
Re: (Score:2)
Simply because it's a *killer feature*. It does things which no native Linux filesystem can offer, even when coupled with LVM and RAID. It takes all of these and replaces them with something that's simply better, more robust and easier to manage. And which is also properly documented with a good number of tutorials and best practices documented clearly.
I'm one of the people who migrated to FreeBSD pretty much because of ZFS, initially for a NAS but now also on other systems. Not the only consideration,
Re: (Score:1)
Isn't FreeBSD the OS where you have to deal with the same incompatibility bullshit as in Linux... just exponentially more so?
Call me pessimistic, but with my luck on Linux I don't even want to try FreeBSD...
Re: (Score:1)
Re: (Score:2)
If I wanted the experience of maintaining a BSD, I'd switch back to Gentoo Linux.
Re: Don't Bother with ZFS (Score:1)
Re: Don't Bother with ZFS (Score:4, Informative)
Re: (Score:2)
It can be argued that testing the OS as a gestalt is a good thing. One good idea is how AIX does releases. You have your OS revisions and your APARs as patches, but you have technology level updates as well, which are not truly OS updates per se... but a snapshot in time where all packages and updates have been thoroughly regression tested. It would be interesting to have something similar on a quarterly basis. This is sort of handled by OS updates like RHEL 6.8 which was recently released, but would be
Re: (Score:1)
It could be worse. I've seen threads dating to around 2001 where the GNAA got into a bot-posting war with the My Clean PC guy (anyone else remember those guys?). Slashdot was basically unusable for an entire weekend as every topic filled up with hundreds of copypasta... ... so, lately its actually been pretty quiet around here, IMHO.
Re: (Score:2)
Re: (Score:2)
Lol, I think the messenger just got shot!
Love ZFS still hoping for BTRFS (Score:2)
I love ZFS on FreeBSD. Its works amazingly well. I tried it on Linux (Antergos lets you use ZFS on root if you're interested). My experience with Linux was less satisfying. It is an absolute memory hog. I was using 8 gigs of RAM at all times. Ordinarily I don't mind this, RAM is there to be used afterall. But on the same box with FreeBSD I rarely broke 2 gigs of RAM used and the same goes with any Linux distro and BTRFS.
The features of ZFS are great and you can't beat the speed and stability but I really ho
Re: (Score:1, Insightful)
Uhm, you know those things are mostly caches? And the seconds it's needed by some program it drops it? Unless you're talking about something like the deduplication support then memory usage is not something you should get to hung up on. I have lots of servers running ZFS, some on as little as 512 MB of RAM and they are fine.
Re: (Score:2)
Re: (Score:2)
Re: (Score:3)
Just create another vdev with the new disks and add it to the pool. Voila, more space in the pool.
Re: (Score:2)
"Creating a ZFS storage pool (zpool) involves making a number of decisions that are relatively permanent because the structure of the pool cannot be changed after the pool has been created. The most important decision is what types of vdevs into which to group the physical disks. See the list of vdev types for details about the possible options. After the pool has been created, most vdev types do not allow additional disks to be added to the vdev. The exceptions are
Re: Love ZFS still hoping for BTRFS (Score:1)
You extend the pool by adding new vdevs, not by adding disks to existing vdevs. I really don't get how that's restrictive.
Re: (Score:2)
No ZFS's thing is snapshots. Once you create a ZFS disk pool I could find no way to add a disk to it to make it bigger.
Add two disks or more :)
Don't think in disks, look up "vdev".
There are a few things to watch out for in linux ZFS for the moment until it catches up. The most important is to label your disks because unlike on solaris, bsd etc if they end up in a different hardware order they will drop out of the pool otherwise.
Re: (Score:2)
Extensible at *which level*?
If you're thinking in terms of filesystem size, and since you mention ext4 this appears to be the case, then this is a complete non-issue. ZFS *datasets* are the equivalent of filesystems, and these can grow and shrink freely; the only limit would be an optional maximum size you set on it, i.e. a quota. Otherwise it use whatever free space is available in the pool. This would be equivalent to growing and shrinking an LVM LV, but without any of the manual effort--to do the equ
Re: (Score:2, Informative)
Linux and BSD accounting for memory currently in use for ZFS ARC differs.
Re: (Score:2)
You have been extraordinarily lucky. No horrible dataloss. And you haven't hit the unblancing issues which make the filesystem read-only after an indeterminate period? Or are you doing a regular rebalance with associated performance loss and hoping for the best?
distributions, not operating systems! (Score:2)
For those not in the known, ZFS on Linux is the official OpenZFS implementation for Linux, which promises to offer native ZFS filesystem support for any Linux distribution, currently supporting Arch Linux, Ubuntu, Fedora, Gentoo, Red Hat Enterprise Linux, CentOS, openSUSE, and now Debian
FTFY
Why, or why not ZFS? (Score:3)
The inclusion of ZFS support into Linux has been discussed for so long I can't remember the first time anymore.
And what I still can't understand is, and please excuse me my ignorance:
1. What is it I can do with ZFS in Linux that is so important?
2. What is it I can't do without ZFS?
I am not saying that we shouldn't support ZFS, because I think we should eventually support anything and everything, but why all the fuss?
Re:Why, or why not ZFS? (Score:5, Informative)
What is it I can do with ZFS in Linux that is so important?
What is it I can't do without ZFS?
It does a lot, but the features I'm interested in are the protection against bit-rot. Specifically, if you set up a mirrored pair of disks in a ZFS pool, it will checksum everything on both sides of the mirror.
When the array is checked (scrubbed), it verifies the checksums. If there is a mismatch because the data has glitched on the media, the checksum won't be valid on one disk, but it will be valid on the other disk so it can repair it. If there's a mismatch in a more conventional mirrored pair, the controller wouldn't really have a way to know which one is correct.
This capability is also in BTRFS, but much has been written about how BTRFS is still experimental. Also, last time I looked, BTRFS was only available for Linux - with ZFS it would be possible to migrate to FreeBSD if Linux does jump the shark.
The other thing is that the scrubbing process is done in the background. My main data pool is a pair of 4TB disks, which was EXT4 to begin with, then BTRFS and now ZFS. The system is a desktop which is powered down at night. Every 180 boots it would run FSCK, which took something like 2 hours to run on EXT4, during which the system was unusable. With BTRFS and ZFS, the scrubbing takes place while the pool is mounted. So yes, you can do this with BTRFS as well, but ZFS is the more proven option of the two.
Re: (Score:2)
Less contraversally it's convenient to have it not only controller agnostic but also OS agnostic. I've connected up a pool created on linux to a FreeBSD box and it just works because it's designed to work in that situation.
Re: (Score:1)
Another company that had a large world to build had issues managing all of the temp files and other crap that goes along with rebuilding a large system entirely from s
Re: (Score:2)
Here's an introduction to ZFS.
* http://wiki.illumos.org/download/attachments/1146951/zfs_last.pdf [zfs]
At the time ZFS was written, nothing even came close to its features.
ZFS got famous for demos like this:
* ZFS is Smashing Baby [youtu.be]
-- :-(
Robin Williams: Lived a hero, died a coward.
Re: (Score:2)
Whoops, that first link should be:
* ZFS: The Last Word In File Systems [illumos.org]
Re: (Score:1)
Good grief, as if the world owes you an explanation - google is your friend! You post is about as useful as people asking for phone numbers of businesses on facebook forums.
- Grumpy Australian
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
No. Because it doesn't violate it. And being a monolithic blob is the least of the criticisms which we could make about systemd, when there's an entire book's worth of bad design in there. ZFS was designed by competent and expert professionals, rather than unprofessional prima donnas, and it shows.
It's a fundamentally different design to traditional UNIX filesystems and disk management, but that doesn't automatically make it a monolithic blob. Is Linux LVM a monolithic blob? That's the level your quest
GPL ? GNU/Linux ? (Score:2)
SFC says ZFS is a GPL violation and "“Almost There” is More Painful Than Proprietary" (see https://sfconservancy.org/blog... [sfconservancy.org] )
If so, surely we need to drop the "GNU" bit, since it is now merely a GNU system over another proprietary (or at least not FOSS, because it is a GPL violation) kernel? Or will rms continue to want crediting for distributions which violate (in his opinion) the very license he created?
Re: (Score:2, Informative)
The SFC is not the be-all end-all legal authority on the GPL. The SFLC, which has been around much longer than the SFC, has a differing opinion on the issue. The only thing for sure is that a resolution will be adjudicated in the courts and not press releases.
Re: (Score:2)
Also, by no means is ZFS proprietary, even the OSF approved CDDL as an open-source license.
Re: (Score:2)
No, it's not a GPL violation. The real problem is ZFS is CDDL licensed, while Linux is GPL
Re: (Score:2)
2. Debian is packaging a ZFS source package which will be compiled on the user's computer. this means a binary form of ZFS will not be distributed by Debian, so there is no license conflict as no binary ZFS module is distributed.
But these days we distribute VMs/AMIs, etc all the time...
3. There is no legal basis for ZFS binary module to be considered a GPL violation. That's just BS being drummed up by the FSF and SFC. Legally speaking, kernel modules are not derivatives of the kernel.
Really? Compilation is not a violation, but we're talking about distribution right?
It's my understanding that the kernel has GPL exceptions for user-space code, but not kernel modules. This is the reasoning behind the horrible nVidia, AMD install nightmares, etc...
Re: (Score:2)
1. I know
2. Doesn't matter - "user does the link" or "user does the compile" is considered subterfuge by rms / FSF, there are endless historical pronouncements on this. See the discussions on gmp/RIPEM for a start (one summary here: http://tech-insider.org/linux/... [tech-insider.org] ). Or any of the old usenet threads on "gpl and plugins".
3. Yes but if the FSF / rms consider it a violation, even if there is no legal basis, then we should drop the GNU/ from Debian etc. in deference to them.
Of course, it is possible that th
Re: (Score:2)
> Who cares? FreeBSD is under a BSD license, which means
> you are really free to do what YOU want,
unless you happen to get your copy from someone who doesn't want you, the user, to have any freedom. In that case, you're fucked - non-copyleft licenses like BSD do not protect YOUR freedom.
With non-copyleft licenses, ANYONE can place ANY restrictions they like on downstream users.
> not what some lunatic's idea of freedom is.
that "lunatic"'s idea of freedom is that ALL users should have the freedom to
Re: (Score:2)
> Oh, please, stop this disinformation once and for all.
How about you stop with your bullshit?
This is NOT about re-licensing. I never said that the BSD-licensed code was somehow magically re-licensed, so you can stop pretending that I did.
However, anyone who distributes BSD-licensed binaries (whether modified or not) is under NO OBLIGATION WHATSOEVER to provide source code to anyone, ever. Not of their modified source and not even of the original unmodified source.
They're free to tell their users "fu
Re: (Score:3)
"The lack of evidence for this claim is evidence of a cover-up to hide the evidence of this claim, and is therefore evidence for this claim" is basically the prime argument for every conspiracy theory argument ever.
Re: This is great news! (Score:1)
Both parent and GP are probably accurate. Most political candidates tell you what you want to hear.