back to article Canonical adds ZFS on root as experimental install option in Ubuntu

Canonical is expanding Ubuntu's support for ZFS, an advanced file system originally developed by Sun Microsystems. Ubuntu's support is based on the ZFS on Linux project, which itself is based on code ported from OpenSolaris, Sun's open-source operating system. It is licensed under Sun's Common Development and Distribution …

Page:

          1. Gordan

            Re: Wow

            Err, no, apparently you didn't, judging by your explanations.

            ZIL turns sync operations into async operations. It is only ever read on an unclean shutdown. It isn't a write cache. Buffering is done in RAM.

            L2ARC is populated by what is evicted from ARC and reset on export/reboot. Unless you have a very long running system with a working set that significantly exceeds your RAM, L2ARC will be we get even populated, let alone used.

            Various things never get cached in ARC (e.g. IIRC sequential reads), because the win is typically not big enough. If it doesn't get cached in ARC, it will never be in L2ARC. So I'll hazard a guess your testing didn't account for anything but the naive case of cat-ing big files to /dev/null.

            1. Kevin McMurtrie Silver badge

              Re: Wow

              I tested it with a file server, mail server, web server, and (legal) Torrent client on a 1Gbps connection. All together there's about 200 to 300GB of live data for a 400GB L2ARC partition. Three days later, the L2ARC was empty and the spinning disks were handling all activity. Doing diagnostics found that ZFS mistakes even moderately large files for uncachable streaming data. I switched off that 'feature.' Now the L2ARC was populating, but only at a few MB/hour. ZFS's default L2ARC population rate throttled to a peak of 8MB/sec. To be exact, it's 8MB/sec * duty cycle - essentially nothing. That has to be tuned too.

              After hours of effort and two days of warm-up, the cache is working well enough that the spinning rust is free to spend time on things that really aren't cachable. I feel like overall the tuning is fragile and shouldn't be trusted.

              The spinning rust is a RAID of the oldest disks I have - 10 year old 2TB WD Green drives, I think. Higher capacity new disks go into the backup system. Older disks taken from the backup system go into the server. It's correct from a storage sizing perspective but backwards for performance. That's why I was looking to ZFS for help.

    1. The Bloke next door

      Re: Wow

      Been running Debian with zfs for root for the past 18 months. Works a treat. But then, have been using zfs since 2005.... I can snapshot before I update ..just in case.

    2. Anonymous Cabbage

      Re: Wow

      FreeNAS's memory requirement comes from neither FreeBSD nor ZFS, but their "interesting" choice of stack layered on top. It's memory-hungry and slow on the server side thanks to being managed through a Django web app, and memory-hungry and slow on the browser side due to the use of one of those dreadful kitchen-sink JavaScript frameworks.

      Saying that, I run FreeNAS 11 on one of those nasty Celeron G1610 Microservers with 4GB of RAM, and it works fine provided I don't need to go near the web interface. Even dedupe works fine on such a small system, provided the recordsize is cranked up to 1M or even 16M so the DDT doesn't get out of hand.

  1. Anonymous Coward
    Anonymous Coward

    Arrogant entitlement

    > The group said that Oracle could "instantly resolve the situation" by relicensing ZFS under GPLv2.

    What?

    You could "instantly resolve the situation" by changing *your* lawyer-friendly license. Your license is the restriction (other opensource projects have no problems), and you want others to change to your rules so you can benefit from their software?

    How would you react if Sony said "We are unable to use Linux in our products due to their licensing restrictions. Linus could instantly resolve the situation by relicensing"

    Anyway, whatever your feelings on the GPL - even if you love it - you should realise that such a stupid statement smacks of irrational fanboism and arrogance, and the PR guy who wrote it should be sacked, for the companies own good.

    1. Anonymous Coward
      Anonymous Coward

      Re: Arrogant entitlement

      The point is that Oracle have never even published a blog post that says that this use of their code is ok. Forget the relicensing would actually be necessary. The original selection of CDDL was expressly to *prevent* ZFS use in Linux.

    2. Anonymous Coward
      Anonymous Coward

      @AC - Re: Arrogant entitlement

      So Sony instead of using the code for free and just obey the rules, they would ask Linus to put a proprietary license on Linux and pay for each copy of the software they will use ? Doesn't look like capitalism to me.

      Gave you a down vote because you fail to notice all licenses are lawyer friendly, not only GPL. I know you meant "developer-friendly" but still....

  2. ChrisPVille
    Devil

    I smell profit

    This is oracle we're talking about. They're probably just waiting for widespread adoption before suing everyone running Linux for having their dubiously licensed kit. I'm sure they will be merciful and allow their victims to buy Oracle Linux licenses as a remedy (one per installed kernel, just to be safe).

    Updated version of their logo ->

    1. TVU Silver badge

      Re: I smell profit

      "This is oracle we're talking about. They're probably just waiting for widespread adoption before suing everyone running Linux for having their dubiously licensed kit"

      These days, Oracle's become a bunch of litigation-hungry lawyers with an IT department.

      Seriously though, proprietary Oracle ZFS and OpenZFS are different beasts and, since 2010 when Oracle ZFS went closed source, increasingly divergent beasts. Before they were unfortunately taken over by Oracle, Sun Microsystems did intentionally put ZFS under an open source CDDL licence so Oracle would be foolish to try to take this matter to court.

      1. Charles 9

        Re: I smell profit

        But here's the billion-dollar question. Why this CDDL and not some other, already-existing license. Did they ever explain just WHY none of the other licenses suited them?

        1. Phil O'Sophical Silver badge

          Re: I smell profit

          Did they ever explain just WHY none of the other licenses suited them?

          https://www.linuxjournal.com/article/9624

        2. jake Silver badge

          Re: I smell profit

          A lot of the license wars boil down to the same as other religions ... egos and angels dancing on pins.

        3. Belperite

          Re: I smell profit

          It may also be stuck in CDDL indefinitely (Oracle's IP hoarding aside) because of this:

          https://www.theregister.co.uk/2010/09/09/oracle_netapp_zfs_dismiss

          I wonder what's in those private dismissal terms?

  3. Jaap Aap

    ''ZFS is widely admired but downsides include high resource requirements and complexity of configuration."

    Don't mess with the configuration unless you're absolutely sure what you are doing. It works great in default configuration. Don't turn on deduplication, etc.

    1. bombastic bob Silver badge
      Devil

      hey - it's NOT that complex you know... just because YOU seem to be afraid to dive under the hood and tinker with it, doesn't mean the REST of the world must behave like a spineless coward.

      however, a cautionary "skim over the manual" is probably wise for ANY new kind of OS toy.

      here's my tweeks:

      vfs.zfs.arc_max="2147483648"

      vfs.zfs.arc_min="536870912"

      Helps a LOT on an older 8Gb RAM workstation. On a 4Gb RAM server box, I use about half those values.

  4. Anonymous Coward
    Anonymous Coward

    yes because I definitely want a system where an accidental arrow-press on the boot loader menu causes the whole system to be obliterated and all my snapshots destroyed

  5. Anonymous Coward
    Anonymous Coward

    I'll stick to the venerable EXT4 and regular back-ups thanks, I don't need ZFS on a desktop or laptop.

    1. hollymcr

      Of course the same arguments applied to EXT4 when it was new too.

      You're wise not to adopt early. But without early adopters (and even now is rather late to be early) it'll never reach the level of stability that has you say the same of ZFS as you say now about EXT4.

      1. jake Silver badge

        However ...

        ... it does not hold that ZFS will get there just because ext4 did.

        1. vtcodger Silver badge

          Re: However ...

          ... it does not hold that ZFS will get there just because ext4 did.

          I believe that ext4 is somewhat backwards compatible with ext2 and ext3. My impression is that ZFS isn't. May not make a difference to those who believe that newer and shinier is always better. But a large fraction of the world operates on the "Whatever can go wrong will go wrong sooner or later" principle. Backward compatibility does matter to them.

      2. Gordan

        Don't know about that, I abandoned ext4 in favour of ZFS close to a decade ago and have been living happily ever since.

    2. Dedobot

      Absolutely.

      In a same way like you don't need cluster on top of desktop's local drives.

      Zfs for NAS and above, desktop's with desktop fs.

    3. rcxb Silver badge

      I don't need ZFS on a desktop or laptop.

      I do... Had a WD drive run fine for years, then it just started silently developing unreadable blocks and corrupting files. At least ZFS would have avoided that SILENTLY part, while RAID-Z would have prevented data loss.

      Still, it's a shame BTRFS has taken so long to develop into a suitable option, and instead RHEL is determined to essentially back-port modern features into XFS (reflinks for batch/offline dedupe, and VDO for compression), and is similarly giving up and adoption ZFS as well.

  6. hmv

    Myth Busting

    Heavy resource utilisation? Yeah, it could be back in the past when Sun launched it and customers tried it out on servers with tiny memory sizes and oodles of spindles. Today? Not so much.

    As to ease of management, it is by far the easiest reasonably capable storage management solution I've used and I've used a fair number (AIX - quite nice, Solaris Disksuite, Veritas Volume manager, something that was branded otherwise but looked a lot like Veritas VM, and a few others).

    1. Jay 2

      Re: Myth Busting

      I recall Solaris DiskSuite being one of the very few applications where it was significantly easier to use the command line to do something opposed to the GUI.

  7. YetAnotherJoeBlow

    Well done

    "To put it in simple terms, GPL is like BSD with one restriction: you are not allowed to attach any more restrictions. Just give it away the same way you received it"

    @AC: Now that is the most succinct definition of the GPL I have ever read; Have an up vote.

  8. CCD

    Running ZFS on Linux and Mac for years without problems

    We've happily based our data storage (but not root filesystems) on ZFS since 2016, although we don't have a huge amount of data. We're not religious about any aspect of software licensing, just endeavour to get our jobs done while staying on the right side of the law.

    Our main server is a 12 year old Dell with 12GB RAM (only a fraction of which is needed) running CentOS + ZFS on Linux, with 2 x 4TB RAID-1 on decent spinning disks (ZFS does the RAID for us). We have an identical backup server to which we zfs send/receive hourly over Gigabit Ethernet and this usually takes a few seconds. With cron we create hourly, daily, weekly and monthly snapshots, with jobs to prune these so we always have access to daily ones for last 10 days, weekly for the last 6 weeks, and monthly ever since we created the pool. We quite often make use of files in historical snapshots. I've managed pairs of NetApp filers, but our current needs are met at a fraction of the cost.

    30 miles away, we have a live offsite server for DR. It's a Raspberry Pi with 1GB RAM running Raspbian Stretch with cheap 2.5" external HD storage to hold the ZFS pool. It gets sent the same hourly streams via OpenVPN over the public Internet, with each sync usually taking just a few minutes, but when it's longer that doesn't matter. If our main office is burnt down, we shouldn't have lost more than an hour's saved work (assuming we get out alive).

    All 3 servers scrub their pool weekly. I've never had a bad block on the main 2 servers, and only had to rebuild the Pi pool a couple of times (bringing it back to the office to sync, which then takes a day or so). On CentOS, with kmod flavour of ZFS modules we don't have to rebuild these for every kernel update - on the Pi rebuild is a bit more tedious, but not required very often.

    Several of our laptops have internal or external ZFS storage, on HD or SSD, using ZFS on Linux or OpenZFS on OSX on macOS (yes, they're compatible if you know what you're doing). These can be re-synced when onsite typically in just a few minutes, even if they're several days out of date. My own laptop runs ZFS on Linux on Debian Stretch.

    My biggest wish is to see CentOS and Debian provide ZFS as standard packages, but I'm not holding my breath. The other area of improvement I'd like is getting the maintainers of the Linux and macOS forks of ZFS to make it easier to create pools without platform-specific additional features that make them non-portable.

    1. sitta_europea Silver badge

      Re: Running ZFS on Linux and Mac for years without problems

      "We've happily based our data storage (but not root filesystems) on ZFS since 2016 [detail snipped]"

      That was very useful. Thank you.

Page:

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like