Stratis Is Red Hat's Plan For Next-Gen Linux Storage Without Btrfs

[duby229]

I've been running btrfs on my home fileserver since 2012. I started with two 2 TB WD Reds. I've added four more drives and replaced the 2 TB drives after one failed. It's up to 30 TB total now in RAID10 configuration.

In all that time it hasn't failed me. It properly corrected all of the bad data when one drive went bad. And it has been great to be able to rebalance to new RAID configurations without shutting down and restoring from backup.

It's nice to have that feature set I'm sure, which is why I'm personally very excited this new initiative exists. Its about time for a real alternative that also meets the same features set. This is a good thing. I'm glad that btrfs is working well for you in that scenario, I hope that stratus can do the same as an alternative.

[gbcox]

I bet if this came from Canonical everyone would be shitting on it.

They should have invested on bcachefs instead.

I'm looking forward to bcachefs but it isn't ready yet - and Redhat is heavily invested in XFS so at this point in time it makes sense.

[mmstick]

> Stratis is developed in Rust and Python.

kthxbye

You'd have to be a brain-dead dingbat to write system-critical software in a non-safe language today. You don't start a new project in 2017 without basing it around Rust. Common sense.

[Anvil]

what's the problem of ext4?

its old an out of date

[jabl]

A couple of days ago RH bought a company called Permabit which makes some block-based deduplication, compression, thin-provisioning software. https://www.redhat.com/en/about/pres...ion-technology

Perhaps another piece of the puzzle to be managed by this stratis thing.

[kreijack]

ext4 lacks many "modern" and advanced features like builtin compression, encryption, RAID, deduplication and more. Even it maintainer (Theodore Ts'o) said that btrfs is future for Linux storage.

The idea behind "stratis", is to add all these feature at block level using new specific DM target. This is philosophically different from the BTRFS/ZFS approach:
- BTRFS/ZFS integrate all the layer (fs, volume manage, raid subsystem): 1 piece of software, highly integrate, very complex to develop and test.
- STRATIS is more modular, this could reduce the develop time and the test time.

The down of the STRATIS approach is that it is more complicated to share the information between the layers: one of the biggest selling point of BTRFS/ZFS is that you can SCRUB/MOVE between the disk only the real data because the filesystem know which zone is used and which not. STRATIS can do it only if the filesystem does a trim/discard of the unused area....

[kreijack]

Seems to be no mentioning of checksumming or how they will deal with problems in all other userspace programs. I will use whatever that solves all problems first but I have a hard time seeing how Stratis can ever solve everything, or even enough things to outweigh the additional problems it will create.

The pdf file stated that the "integrity checking" will be developed as DM target (see cap 10.2.2); it will appear in the rev 3.0

[dfyt]

Ext4 is currently the best filesystem for normal users since its very stable and is the fastest overall. But for the future they need many new features that aren't part of ext4 design (and also great performance, maybe thats why btrfs is not an option anymore), so this may be a great idea.

My experience has been the opposite, having used Linux since 96, I switched over from EXT to XFS about 4 years ago. It's faster, more efficient (I can bit 2TB of my images on my 2TB drive vs 1.7TB). On my servers I use ZFS but desktops XFS and LVM+XFS of multi drive desktops. BTRFS was just pathetically unstable with reboots or crashes and that's not even in RAID config. I think XFS+LVM is the right solution.

[starshipeleven]

EXT4 never fails to me, have a good performance and works well with ssd why they don't invests more in EXT4, making another file system is the solution? duplicating work, if was Canonical doing this, I can image all the hatters here

RedHat has traditionally invested in XFS since a long time ago and they won't change that now, since they are already in XFS why not develop on it this new add-on system.

Really it's not duplication, it's "how about we add features we need to stuff we know best".

[Gymnasiast]

I bet if this came from Canonical everyone would be shitting on it.

They should have invested on bcachefs instead.

My thoughts exactly. Diversity and choice are something to be celebrated in Linux land, unless it's developed by Canonical, it appears.

I do wonder where Red Hat's priorities lie. They develop another next-gen filesystem, yet they don't tackle one of the biggest problems by giving Mr Poettering a damn good caning. (It's no good having the attitude of a Torvalds or de Raadt without the accompanying talent.)