OpenZFS Eyes Faster Scrub, Improved Compression, uZFS, Better Performance

[phoronix]

Phoronix: OpenZFS Eyes Faster Scrub, Improved Compression, uZFS, Better Performance

Taking place yesterday and today in San Francisco has been the annual OpenZFS Developer Summit. Talks this year ranged from how Amazon AWS is making use of OpenZFS to a number of optimizations and improvements currently being tackled by open-source developers...

https://www.phoronix.com/news/OpenZFS-Dev-Summit-2022

[mrtruxton]

I find myself doing more and more with ZFS.

It started with Proxmox on the hypervisor level. We also started using it inside VMs so our backup solution could make point in time snapshots.
It does reduce performance a bit but in the NVME age that is not a big issue for the loads we have. ( mostly CPU intensive )

We then started doing Kubernetes on hardware and OpenEBS with ZFS Local PV ( which uses ZFS as a backend: https://github.com/openebs/zfs-localpv )

I think the OpenEBS Cstor backend uses uZFS for creating clustered kubernetes storage if you need that.

https://github.com/openebs/cstor-operators/blob/develop/docs/quick.md

We don't use this outselves but it looks interesting as a concept. ( not sure i would rely on this for my production data t.b.h. )

Its great to see that despite the age of ZFS its still alive and kicking and the foundation of many new solutions.

[skeevy420]

ZFS is my go-to for non-root volumes that aren't portable. For that role I don't see the point of other file systems.

A few months ago I was looking at how old ZFS's LZ4 was in comparison to upstream so that was good news to read. Decompression improvements are always a good thing.

Block cloning sounds very interesting and I hope it works as well multiple Wine prefixes as I'm hoping it will.

[reza]

As one of the advantages/motivations of uZFS is ease of development on user-base, I'm curious to know what people think of Stralis? Stralis has that aim as well among other things.

[dragonn]

I would love to see ZFS vs uZFS vs ZFS-fuse benchmarks!

[darkbasic]

I'm curious to know what people think of Stralis? Stralis has that aim as well among other things.

You mean Stratis?

[skeevy420]

I'm curious to know what people think of Stralis? Stralis has that aim as well among other things.

It's called Stratis.

A lot of us were hyped when it first came out, but the more we've had time to think about it the less we want to use it. It doesn't do anything better than BTRFS or ZFS and the things it does do it does worse than them and will never surpass them. I used to be gung-ho about it thinking we'll have a GPL equivalent of ZFS, but 4 or 5 years later I see that Stratis will never be that.

Stratis is basically a fustercluck of XFS on top of LVM/device-mapper on top of LUKS/Clevis. Instead of following the UNIX principle of using one tool to do the job right it's a bunch of tools glued together. The file system is one tool, pools and volumes are other tools, and encryption are even more tools. Stratis is exclusively Linux.

BTRFS does all of that Stratis does with just BTRFS on top of LVM (BTRFS isn't the greatest with nested subvolumes and random pools so it needs LVM). BTRFS also has the potential to do all of that with just BTRFS. It just needs better pool and subvolume handling. Pool and subvolume handling is where ZFS kicks ass and really, really shines. BTRFS has the most potential to be the most UNIX-principled file system in the kernel. BTRFS has a Windows port that's almost feature complete.

ZFS does all that BTRFS and Stratis do with just ZFS making it the most UNIX-principled file system out there. It's also the most multiplatform of the bunch and runs on Linux, FreeBSD, Solaris, and macOS with Windows support coming in the near future.

[reza]

Yeah sorry for the typo, I meant Stratis.

I like BTRFS even as my daily driver. I also use WinBTRFS but my issue is Quibble which is very buggy and I have tried many times and never been able to boot from BTRFS root partition (https://github.com/maharmstone/quibble/issues/66)

[pepoluan]

Its great to see that despite the age of ZFS its still alive and kicking and the foundation of many new solutions.

Rather than saying "ZFS is an aged filesystem" you really should be saying "ZFS is a field-tested filesystem, with more than a decade of Production use and optimisations."

Because ZFS of today is leaps and bounds better than ZFS of yore; it's not a static thing, it keeps being improved and extended with one overarching principle: Integrity of your data is the utmost importance.

[evil_core]

Rather than saying "ZFS is an aged filesystem" you really should be saying "ZFS is a field-tested filesystem, with more than a decade of Production use and optimisations."

Because ZFS of today is leaps and bounds better than ZFS of yore; it's not a static thing, it keeps being improved and extended with one overarching principle: Integrity of your data is the utmost importance.

But it shows it's age and limits.
- It's slow
- You can not defragment it(and there are no plans. BTRFS got it from years)
- you can not 'rebalance' it, etc, like BTRFS(or LVM)
- it's not tiered and doesn't support fancy caching, like BcacheFS.

So while ZFS is not dead yet, I believe it will be replaced by BCacheFS(or even something better) in next decade or two.