Learning More About Red Hat's Stratis Project To Offer Btrfs/ZFS-Like Functionality

[pipe13]

Ah... the perpetual beta-ness that is Fedora! There will supposedly be a Stratis 0.5 release for Fedora 28. From https://fedoraproject.org/wiki/Changes/StratisStorage:

"Until version 1.0, Stratis should not be used on a system with valuable data. Stratis will not be well-integrated into the boot process, nor will placing root partition on it be possible. However, Fedora users and testers can try Stratis on non-primary disks, and put it through its paces to generate bug reports and feature requests that can guide Stratis development leading up to a stable release."

Actually, there's probably no better way to put some miles on this thing. It's not like it will be installed by default: folks who want it will have to deliberately "dnf install stratis".

[Zan Lynx]

Its performance is also utterly abysmal as evidenced by regular benchmarks here on Phoronix, far behind XFS on the same work loads.

The fair way to compare BTRFS and XFS would be BTRFS against XFS on top of LVM snapshot. Watch XFS performance dive as LVM frantically tries to copy every written block into the snapshot set.

[salsadoom]

Well... maybe this isn't a bad idea. My first reaction "not another one...!" but at the same time, this isn't actually entirely from scratch. It seems mostly to be combining a lot of existing, tested solutions into a single package. That's a good idea, since we've been waiting for btrfs for a long time and I don't see that changing soon. Code reuse is a good thing, especially when its stable and trusted code. XFS is good stuff, so... I donno, I'm not entirely displeased with the existence of the project, I'll just wait and see what actually happens.

[F.Ultra]

Actually, the problem with BTRFS is Oracle. Yes BTRFS is part of the kernel, and yes there are others helping with it, but the fact remains that it's generally supported and steered by Oracle. Its performance is also utterly abysmal as evidenced by regular benchmarks here on Phoronix, far behind XFS on the same work loads. Comparing BTRFS against OpenZFS is pointless because ZFS can't be integrated into Linux without a clean room project. The only reason you'd want to compare them as a user is when evaluating Linux versus FreeBSD servers rather than Linux distributions among themselves.

Frankly, only a fool would trust Oracle for anything, especially their direct competitor Red Hat. Oracle is a treacherous company. If Red Hat wanted to work on BTRFS based projects they could hire on developers if they needed to. I personally believe Red Hat would be foolish to do so.

BTRFS might have started out as a Oracle project but since 2013 the main lead (Chris Mason) works at Facebook, if that makes the situation better or worse is up to debate of course . The performance penalty however is something that you have to pay if you want to have strong checksumming on every read and write (and a Copy On Write design so that you can create instantaneous snapshots). It's a trade-off between performance and features, put the same features into XFS and it will be just as slow (and since XFS does not employ a COW design you cannot make it feature complete anyway).

[andyprough]

Its performance is also utterly abysmal as evidenced by regular benchmarks here on Phoronix, far behind XFS on the same work loads. Comparing BTRFS against OpenZFS is pointless because ZFS can't be integrated into Linux without a clean room project. The only reason you'd want to compare them as a user is when evaluating Linux versus FreeBSD servers rather than Linux distributions among themselves.

I've run it on quite a few office computers and never had a performance problem. I guess if your whole world revolves around video game performance, then you'd probably have a big concern.

[GizmoChicken]

At one time, XFS was working to provide copy-on-write (CoW) and online scrubbing functionalities. I don't closely follow XFS, so sorry if I missed this, but what's the current status of those functionalities in XFS?

I ask because I've been using BTRFS for a few years now, but not because of its volume-managing filesystem (VMF) functionality. Rather, I use BTRFS for its CoW and scrubbing functionalities.

If XFS doesn't (or won't soon) offer CoW and scrubbing functionalities, I wonder why Stratis isn't being build on top of a file system that does. And thoughts on this?

[dwagner]

The fair way to compare BTRFS and XFS would be BTRFS against XFS on top of LVM snapshot. Watch XFS performance dive as LVM frantically tries to copy every written block into the snapshot set.

Exactly!

I use both XFS and btrfs, and XFS is clearly the better option if you do not need snapshots. But if you do - for example, because you need to perform online backups of filesystems that are used 27/7, the btrfs retains reasonable performance, while XFS on a LVM with a snapshot crawls into basically becoming "unavailable for writing".

I don't see how RedHat's "Stratis" is something new. It looks and smells like XFS on LVM, and will probably show the same performance characteristics.

Maybe btrfs does have some serious design flaws, maybe it could be much better, but unless someone presents an actually better performing filesystem for writing while a consistent snapshot exists, I don't assign any blame to the btrfs developers.

[dwagner]

At one time, XFS was working to provide copy-on-write (CoW) and online scrubbing functionalities. I don't closely follow XFS, so sorry if I missed this, but what's the current status of those functionalities in XFS?

AFAIK, they are still in development. But I do not hold my breath until that is finished, because once it his, who says it will perform much better than btrfs when COW / snapshots are in use?

[starshipeleven]

Actually, the problem with BTRFS is Oracle.

I believe that most of the development for BTRFS is now being done at Facebook.

Wrong for both.

It's SUSE the leading force on btrfs, then there is Oracle and Fujitsu plus some oddballs here and there. Facebook was active in adding stuff like their new compression algorithm, but isn't terribly active on other topics.

[starshipeleven]

Ah... the perpetual beta-ness that is Fedora! There will supposedly be a Stratis 0.5 release for Fedora 28. From https://fedoraproject.org/wiki/Changes/StratisStorage:

"Until version 1.0, Stratis should not be used on a system with valuable data. Stratis will not be well-integrated into the boot process, nor will placing root partition on it be possible. However, Fedora users and testers can try Stratis on non-primary disks, and put it through its paces to generate bug reports and feature requests that can guide Stratis development leading up to a stable release."

Actually, there's probably no better way to put some miles on this thing. It's not like it will be installed by default: folks who want it will have to deliberately "dnf install stratis".

Perpetual Beta OS gets beta "filesystem API" feature for voluntary testers, more at 20:00.