Canonical Developers Preparing For More ZFS Improvements In Ubuntu 20.10

[k1e0x]

Would love to see a ZFS on root setup as easy as the FreeBSD installer makes it. Just about anyone smart enough to install an OS can install an encrypted ZFS with Raid 1 or 0 if they want it just using the FreeBSD installer. NetBSD 10 might have ZFS on root bootable too. Coupled with the Illuminos distros that and assuming Ubuntu gets its stuff in order that will be four easy to roll out ZFS on root OSes.

Baby steps. Ubuntu has a lot to get right and hasn't had 10 years of ZFS integration like FreeBSD has had. Supposedly macOS will boot on ZFS as well so you can add that to your list. I've used ZFS volumes on macOS but not as a system boot volume (root).

Yeah, because OpenZFS isn't open source...
Seriously, where do you people come up with this crap?

Where do you think? Britoid well knows ZFS is OpenSource and more permissively licensed than Btrfs. He just doesn't want to admit that and wants to spread FUD for some reason or another..

[Britoid]

Yeah, because OpenZFS isn't open source...
Seriously, where do you people come up with this crap?

Java is open source too. At least, that's what Google thought.

[DanL]

Java is open source too. At least, that's what Google thought.

Huh? Google thought they could make their own implementation of Java API('s) without having access to the source code. They didn't think JavaSE was open source, and even tried to buy a license for it before writing code.

That's very different than the Canonical/OpenZFS situation. OpenZFS is indeed open source, licensed under CDDL.

[Mario Junior]

Everyone have basically given up on BTRFS for desktop distros. While Canonical went with ZFS, Red Hat is trying to develop similar features over XFS instead.

Hopefully BcacheFS can eventually deliver a good CoW FS to linux but till then everyone just have to make do with what came before.

Not SUSE.

[Mario Junior]

Doesn't openSUSE use BtrFS by default? I'm actually considering jumping ship there from Kubuntu, because I don't like the progressing snapification and I already have my data partition as BtrFS.

Yes and is the best BTRFS implementation.

Doesn't openSUSE put /home on XFS still though?

Nope. Is BTRFS on all partitions (except EFI, of course).

If you're not on OpenSUSE, join us... the btrfs is warm

Already testing

UUID=ea9ec070-e644-4067-8bfe-b96b6579830d / btrfs compress=zstd,ssd_spread,noatime,space_cache 0 0
UUID=ea9ec070-e644-4067-8bfe-b96b6579830d /.snapshots btrfs subvol=/@/.snapshots 0 0
UUID=ea9ec070-e644-4067-8bfe-b96b6579830d /var btrfs subvol=/@/var 0 0
UUID=ea9ec070-e644-4067-8bfe-b96b6579830d /usr/local btrfs subvol=/@/usr/local 0 0
UUID=ea9ec070-e644-4067-8bfe-b96b6579830d /tmp btrfs subvol=/@/tmp 0 0
UUID=ea9ec070-e644-4067-8bfe-b96b6579830d /srv btrfs subvol=/@/srv 0 0
UUID=ea9ec070-e644-4067-8bfe-b96b6579830d /root btrfs subvol=/@/root 0 0
UUID=ea9ec070-e644-4067-8bfe-b96b6579830d /opt btrfs subvol=/@/opt 0 0
UUID=ea9ec070-e644-4067-8bfe-b96b6579830d /home btrfs subvol=/@/home compress=zstd,ssd_spread,noatime,space_cache 0 0
UUID=ea9ec070-e644-4067-8bfe-b96b6579830d /boot/grub2/x86_64-efi btrfs subvol=/@/boot/grub2/x86_64-efi 0 0
UUID=ea9ec070-e644-4067-8bfe-b96b6579830d /boot/grub2/i386-pc btrfs subvol=/@/boot/grub2/i386-pc 0 0
UUID=288D-4F65 /boot/efi vfat utf8 0 2

[flower]

I run Btrfs as the root filesystem on all of my Linux desktops, and I'd say it's a really solid filesystem as long as you know how to use it.

IMHO if you need to have special knowledge to use a filesystem that filesystems just sucks.

i dont even see a reason to use it. except for offline(!) dedup and data checksumming all features are available through lvm/ext4 (even dedup)

i used btrfs for a while and it worked well except for my vm's which got REALLY slow.
i know i could have disabled cow for them - but i used snapshots for backups.

now i can make a lvm snapshot for backups and my vm's just work.
that's how a filesystem should work: you install and forget it

[Chugworth]

IMHO if you need to have special knowledge to use a filesystem that filesystems just sucks.

i dont even see a reason to use it. except for offline(!) dedup and data checksumming all features are available through lvm/ext4 (even dedup)

i used btrfs for a while and it worked well except for my vm's which got REALLY slow.
i know i could have disabled cow for them - but i used snapshots for backups.

now i can make a lvm snapshot for backups and my vm's just work.
that's how a filesystem should work: you install and forget it

Keeping system snapshots is a very handy feature, and neither Btrfs or ZFS should have the snapshot performance impact of LVM. Both filesystems were designed from the ground up with CoW support. In fact, I would expect the performance of LVM to be as bad or worse than Btrfs when it comes to keeping snapshots of VMs. In that case, ZFS would be the best choice due to the way it caches data, and also for its ZVOL feature. But you would really need a lot of memory for it to work ideally.

[flower]

Keeping system snapshots is a very handy feature, and neither Btrfs or ZFS should have the snapshot performance impact of LVM. Both filesystems were designed from the ground up with CoW support. In fact, I would expect the performance of LVM to be as bad or worse than Btrfs when it comes to keeping snapshots of VMs. In that case, ZFS would be the best choice due to the way it caches data, and also for its ZVOL feature. But you would really need a lot of memory for it to work ideally.

the problem with VMs is fragmentation (same for db - eg your firefox/thunderbird profile is a sqlite db).
usually my win10 vm starts up in 15secs - with lot of frag it got down to 5 minutes!

[skeevy420]

Michael

Sorry, I fixed a typo...

[Chugworth]

the problem with VMs is fragmentation (same for db - eg your firefox/thunderbird profile is a sqlite db).
usually my win10 vm starts up in 15secs - with lot of frag it got down to 5 minutes!

Right, any CoW filesystem is going to suffer from fragmentation. ZFS deals with it by caching the most frequently accessed data at the block level rather than the file level. That's why a large amount of memory would help - enough to allow plenty for the VM's and also plenty for the ZFS cache. You could also add an SSD drive to the pool that's used for nothing but caching.