Approved: Fedora 33 Desktop Variants Defaulting To Btrfs File-System

[k1e0x]

No, I only want to say that what btrfs does is still "online deduplication" and you should really stop moving goalposts.

I run ZFS in my KVM server (VMs virtual drives are vdevs, all managed by libvirt/Virt Manager), and I did try dedupe, yes. It's an overrated feature, not for ZFS issues mind me, it's just that deduplication in general is amazing at some specific niche cases and very meh at most other cases. I really don't have dozens of same-y VMs with just a few changes, so it's mostly useless for me.

I would rather have a defrag tool, like btrfs has, and not having to rely on send/receive (to another filesystem and back) to "defrag" a vdev.

Should try it the other way around. ZVOL's make excellent vm disks. (a lot better than qcow, qcow2, etc) it's also supported in libvirt but in virt-manager there is a bug that detects the datasets incorrectly and you have to manually specify the path.. (/dev/zvol/pool/dataset/disk) otherwise it works good.

The dedup in ZFS now really isn't so hot. It's mostly useful for maybe similar blocks on a VM cluster or a directory full of hash files, other than that.. mmm.. It loads that entire bitmap into ram right now and it does so forever. Keep in mind that everything ZFS does is based on blocks, not files so dedup works on blocks.. there are some advantages to that (such as VM's) but it isn't what a lot of end users expect. I would caution ppl know exactly what they are doing before using it. I would expect the new version to be maybe a year or two out. (if ZFS encryption implementation timeline is an example.. big project and lot of work and it took about 2 years for us to get that.)

[pal666]

??? So what you are saying is.. if I format my disk with BTRFS, and then I smash it to bits with a hammer.. BTRFS still has my files?

no, that's your imagination. what i'm saying is bitrot happens on single devices too and extra copy with scrub will autoheal

[pal666]

: face palm :

ZFS does the same thing. I wonder where BTRFS got the idea. heh..

so zfs has same thing but with excessive memory usage. ok

[pal666]

Thank you for the confirmation that only way to have dedup on btrfs is offline

it's online, it's just not during writes, but after writes

[kloczek]

No, I only want to say that what btrfs does is still "online deduplication" and you should really stop moving goalposts.

And I already wrot i that without keeping enough number of hashes of blocks in memory btrfs dedup is useless (is only technology demonstration).

I run ZFS in my KVM server (VMs virtual drives are vdevs, all managed by libvirt/Virt Manager), and I did try dedupe, yes. It's an overrated feature, not for ZFS issues mind me, it's just that deduplication in general is amazing at some specific niche cases and very meh at most other cases. I really don't have dozens of same-y VMs with just a few changes, so it's mostly useless for me.

I would rather have a defrag tool, like btrfs has, and not having to rely on send/receive (to another filesystem and back) to "defrag" a vdev.

On cloning volimes you don't need to dedup anything.,
That type of dedup was availaible on ZFS since it was born. That is immanent part of any fs which has cloces (RW snapshots).
Defrag is not needed.
Look on "zdb -vvv" output and you will find SLAB metrics data.
SLAB allocator is probe of any fragmentation.
Please read SLAB allocator documentaion (quite good is availabe on Linux).

btrfs is not usinfg SLAB allocator and this is why it needs and can be defragmented.

[kloczek]

btrfs balance start -dconvert=dup /mnt
moron, btrfs supported it from the start, zfs added it in 2010

Wrong. ZFS even today does not have such thing becase it does not need it

[k1e0x]

pal666

Just go out and play with it some, I'm sure you'll find a lot of the same features. (ZFS came first and invented many of the features after all) - If you're all concerned about this "out of kernel" thing. Use it on FreeBSD, it's in the kernel there. It makes one hell of a powerful storage platform.

You know ZFS send can do a delta copy of an encrypted dataset? Pretty cool. I think BTRFS still uses LUKS right?

A use case for that could be.. you want to push a copy of the dataset into the cloud but you want to keep it private and you don't want to spend and hour calculating the delta like with rsync. ZFS already knows the difference between blocks so all it has to do is send and it does not need to send the private key. That Tom Caputi guy that wrote that is pretty clever..

[kloczek]

it's online, it's just not during writes, but after writes

Thank you for confirmtion second time that dedup is not onliine and is offline.
Loook .. proper dedup domne online is able to lower write IOs
What has btrfs only os able to lower used disk space.
Do you see the difference between dedup and pseudo dedup which has btrfs?

[starshipeleven]

Deftragmentation is completly not needed.

bullshit https://forum.proxmox.com/threads/ho...fs-pool.42931/

https://github.com/openzfs/zfs/issues/3582

https://github.com/salesforce/zfs_defrag

Did you heard that someone is doing "RAM defrgmentation" on Linux?

https://www.kernel.org/doc/html/late...compact-memory
echo 1 > /proc/sys/vm/compact_memory
It's not commonly needed as only some specific workloads need it, but yes sometimes it's necessary.

[pal666]

Oracle doesn't own OpenZFS.

oracle owns oracle zfs, i.e. the original stable and enterprise-grade filesystem, not some noname garage production

Oracle does have OracleZFS but that version was never ported

if you were programmer, you would know that by porting filesystem to different os you don't get same filesystem, you get something completely different, which inherits neither speed nor correctness, that you'll have to build from scratch

The problem really isn't with the CDDL

as i said, i don't care. license choice is the problem of code owner, and he has to live with it

Distros can bundle it and support it.. just like Ubuntu has done.

only distros who have nothing to lose

Do you own an Nvidia card?

owning most anti-linux card would be stupid, wouldn't it?