Recommend a filesystem for SSD

[pjp]

I can't think of why filesystems could be that exciting.

As a systems admin, it made life a lot easier.

I don't understand how the ZFSoL kernel modules and the kernel work together or I might use it. Loosely I understand that whenever the kernel is upgraded, I think the modules need to be rebuilt. That doesn't seem too bad. But I've read of enough problems that I just don't want the headache. On the other hand, I still don't have a backup / "NAS" system built either.

[logrusx]

I just don't want the headache.

Same here.

Best Regards,
Georgi

[lars_the_bear]

But I've read of enough problems that I just don't want the headache. On the other hand, I still don't have a backup / "NAS" system built either.

I suspect that most of the features of ZFS that made it appealing back in the day are now handled by other parts of Linux. LVM handles the storage pooling stuff and (to some extent) snapshots. Linux software RAID handles the RAID stuff. Some other Linux filesystems support copy-on-write semantics.

I use ext4 on my NAS, and I've never really had any reason to complain. I'm aware of its notional lack of features compared to more modern implementations, but it's done fine for me.

BR, Lars.

[pjp]

I don't know when Linux gained reliable LVM and mdraid support, but those options were used before ZFS. ZFS unifying those capabilities into a filesystem and tools makes it a lot easier / cleaner / simpler. I was disappointed when I learned bcachefs modeled it's CLI after btrfs. The btrfs CLI design principle seemed to be "Not like ZFS," which is a shame, because btrfs seems convoluted by comparison.

[steve_v]

I didn't see any easy way to know which kernel they support.

Github -> Releases, top of each release listing, e.g.

zfs-2.2.5 Latest
Supported Platforms

Linux: compatible with 4.18 - 6.9 kernels
FreeBSD: compatible with releases starting from 12.2-RELEASE

ZFS itself is good. but I'm not convinced about ZFSoL. I'd use it without hesitation on BSD.

Been running a Debian/Devuan fileserver with ZFS as the main storage pool since 2013, no problems to report.
Can't comment on life with ZFS on Gentoo or other rolling-release distros, but I'd expect at least some aggravation same as any other out-of-tree module - i.e. the usual compatibility check prior to kernel upgrade and module rebuild after.

I suspect that most of the features of ZFS that made it appealing back in the day are now handled by other parts of Linux. LVM handles the storage pooling stuff and (to some extent) snapshots. Linux software RAID handles the RAID stuff. Some other Linux filesystems support copy-on-write semantics.

Full hierarchical checksums with seamless automatic correction? Transparent compression and/or encryption? Multiple file copies? Virtual raw block devices (zvol)? Adaptive read cache with dedicated cache and/or write-intent devices? Deduplication? Online transaction-based network replication? Excellent management tools with consistent and non-aggravating syntax?
Linux native solutions are indeed improving, but EXT4 on MDRAID + LVM is hardly comparable to ZFS.

All that said, to return to the OP I wouldn't run ZFS on a single SSD, that's not what it's for.

[lars_the_bear]

Full hierarchical checksums with seamless automatic correction? Transparent compression and/or encryption? Multiple file copies? Virtual raw block devices (zvol)? Adaptive read cache with dedicated cache and/or write-intent devices? Deduplication? Online transaction-based network replication? Excellent management tools with consistent and non-aggravating syntax?
Linux native solutions are indeed improving, but EXT4 on MDRAID + LVM is hardly comparable to ZFS.

You're preaching to the choir I like ZFS. At least, I liked it on Solaris. I haven't used it on Linux beyond the 'ooh, look -- it works on Linux' point, because I don't have enough need for what it offers, to go to the hassle; and because it's not supported by Dropbox.

BR, Lars.

[pjp]

Github -> Releases, top of each release listing, e.g.

zfs-2.2.5 Latest
Supported Platforms

Linux: compatible with 4.18 - 6.9 kernels
FreeBSD: compatible with releases starting from 12.2-RELEASE

Thanks, but that kind of emphasizes the point. When I got to zfsonlinux.org, I expect to find basic requirements information. Especially given that it is a critical component of updates. I don't see any links that facilitate discovering that information. Maybe it's an intentional choice.

ZFS itself is good. but I'm not convinced about ZFSoL. I'd use it without hesitation on BSD.

Been running a Debian/Devuan fileserver with ZFS as the main storage pool since 2013, no problems to report.
Can't comment on life with ZFS on Gentoo or other rolling-release distros, but I'd expect at least some aggravation same as any other out-of-tree module - i.e. the usual compatibility check prior to kernel upgrade and module rebuild after.

I'm only basing that on issues I've seen here. And at least some point, there seemed to be issues / lack of information on how to correctly configure it for Gentoo. I don't recall specifics, but I'm sure it would turn up with a search (I don't think there are that many topics).

All that said, to return to the OP I wouldn't run ZFS on a single SSD, that's not what it's for.

Depends on why you think "that's not what it's for." A long time ago I believe it was a ZFS developer who wrote about recovering a photograph that had been corrupted. Seems like a good use for it to me. I can't find it now, but I was thinking it was an ars article, but maybe it was on one of the Sun blogs.

[steve_v]

Depends on why you think "that's not what it's for."

I mean ZFS is, or at least was historically, an enterprise filesystem intended to be run on "big" Sun systems with many disks. A single device pool still gives you checksums, snapshots, etc. of course but no real redundancy or recovery options (i.e. no fsck) if the device fails hard enough to corrupt the filesystem.
IMO, if you're not leveraging the multi-device side of ZFS/ZoL, you're probably better off using something that doesn't have the hassle of out-of-tree non-gpl modules, like BTRFS.

[pjp]

I mean ZFS is, or at least was historically, an enterprise filesystem intended to be run on "big" Sun systems with many disks. A single device pool still gives you checksums, snapshots, etc. of course but no real redundancy or recovery options (i.e. no fsck) if the device fails hard enough to corrupt the filesystem.

I don't consider origianl usage a limiting factor since other benefits are a big improvement over standard alternatives. Similarly with a separate /usr and the discovered benefit of mounting it read only.

I'd forgotten about fsck. I'll have to look into and consider that. zfs-send / zfs-receive may outweigh the drawbacks. If I'm not mistaken, checksums achieve something similar to journaling. So single disk failure doesn't seem any worse. Unless I'm missing something.

IMO, if you're not leveraging the multi-device side of ZFS/ZoL, you're probably better off using something that doesn't have the hassle of out-of-tree non-gpl modules, like BTRFS.

Can the kernel no longer be patched with ZFS? I thought the out-of-tree modules were a workaround to shipping a kernel / binary distro with ZFS by default. Managing upgrade boot environments is one of the primary benefits, so the out-of-tree modules makes it much less compelling.

The hassle of not-ZFS is the reason I'm considering it. I'll have to look at requirements and see if I can do some VM testing.

LVM snapshots are less effective if /boot is separate, so I'll have to see how much that would limit ZFS. Although I guess nothing works when EFI is involved, and my laptop is EFI based.

Who could have guessed there are very good reasons for separate /bin and /usr.


As an aside, I found the article I mentioned. I was off on the details. It was from 2014 and the author wasn't a ZFS dev. It was about standard protection of multiple drives with ZFS and btrfs.

Nevertheless, I believe comparable "bit rot" protection can be had with a single drive and multiple metadata copies. Obviously at the cost of some additional storage. That doesn't seem worse than non-ZFS solutions, plus the other very useful features.

https://arstechnica.com/information-tec ... lesystems/

[turtles]

All that said, to return to the OP I wouldn't run ZFS on a single SSD, that's not what it's for.

Good info

I was involved in the thread where somebody complained about kernels 6.9 not being in tree anymore and they can't use ZFS because of that.

ooouch that sounds like a rough pothole to avoid.

What filesystem do you guys think is the best for a single SSD Gentoo (or any Linux distro) laptop?

[logrusx]

I don't think it matters much for a single SSD and desktop usage.

[NeddySeagoon]

turtles,

From a FLASH wear perspective, with a modern SSD, you will replace it before it wears out.

From habit I use ext4 everywhere but where things are local caches, like distfiles, repos and build space which are all disponable, I don't bother with a journal, since the recovery process is to throw it away and make anew filesystem.

I expect my SSDs to outlast me anyway ... so like I say, it's from habit, its not required to avoid wear,

[nikolis]

ext4 on /.
Gives a simple, mature, and highly predictable base for the operating system.

btrfs on /home.
Adds snapshots, subvolumes, and more flexible backup/rollback options for personal data.

This way, the system stays stable and low-maintenance, while the home partition remains more flexible and resilient.