A Quick Look At EXT4 vs. ZFS Performance On Ubuntu 19.10 With An NVMe SSD

[S.Pam]

But is periodic "scrubbing" a "good" answer? Maybe it's "good enough"? Most RAID subsystems do periodic scrubbing.

No. Not really.

[cjcox]

No. Not really.

AFAIK ReFS does periodic scrubbing as its technique.

[jacob]

Quite frankly am I the only one who doesn't share this fascination with ZFS? Its performance is absolutely dreadful, its integration into Linux shaky at best (both technically and licence-wise) and we don't even have any figures of its impact on battery autonomy, which I imagine will be major. Sure, CoW is a desirable feature but we already have it in byrfs which, for all its own flaws, is a better and more capable design and I wish efforts were spent on giving it the one feature it's missing, namely subvolume encryption, rather than importing this hodgepodge of enterprisey bloatware.

[johnc]

Quite frankly am I the only one who doesn't share this fascination with ZFS? Its performance is absolutely dreadful, its integration into Linux shaky at best (both technically and licence-wise) and we don't even have any figures of its impact on battery autonomy, which I imagine will be major. Sure, CoW is a desirable feature but we already have it in byrfs which, for all its own flaws, is a better and more capable design and I wish efforts were spent on giving it the one feature it's missing, namely subvolume encryption, rather than importing this hodgepodge of enterprisey bloatware.

It has its uses. The people working on ZFS want it for their use cases, so that's why there is work being done there. If people wanted btrfs, they would be working on that.

[carewolf]

XFS always worked well for me... plain, simple and fast... ZFS looks cool but with a very big performance penalty for now...

ZFS is among the fastest filesystems, and probably beats XFS is every category. Ext4 is just top dog.

[carewolf]

Mostly agree. Though EXT4, LVM and MD RAID does not protect or detect bit rot. So if your data is valuable... Btrfs or ZFS is the way to go. Backups do not help against bit rot since you usually don't detect them before the rot is copied into the backups.

BTRFS will just eat all your data instead of a single file if it hits bitrot in its metadata. Not really an improvement.

[pkese]

ZFS is among the fastest filesystems, and probably beats XFS is every category.

... except that benchmarks prove you wrong.

[nikodll]

It would made a more fair comparison when some features were enabled for both systems. E.g. ZFS + native encryption vs. EXT4 + LUKS.

[adanisch]

I agree with the comments related to ZFS not being able to shine in a one-disk setup. Here is some test data related to ZFS with multiple disk set ups, next time maybe test against ext4 using a RAID10 or something along those lines. https://calomel.org/zfs_raid_speed_capacity.html

[polarathene]

Just me, but IMHO, for ZFS, you really need the muitiple disk aspect of it.

That's less flexible compared to BTRFS right? Besides flexibility in capacity, you can't as easily use it straight away or something? I don't know much about ZFS, just recall that the more disks, the longer it takes to expand a pool with new storage?

The other is that for some strange reason some of my GOG games, and I don't remember which ones, would inexplicably crash when I was using XFS for the drive they were installed on. No effin clue why, but changing it to ext4 and all was fine.

I recall similar issues with Steam few years back(though they've fixed it since iirc. Was something like XFS using inode64 by default, and that caused some problems with data access by certain software some how. There may have been some other issue like that where EXT4 works but XFS doesn't, can't recall.

Quite frankly am I the only one who doesn't share this fascination with ZFS? Its performance is absolutely dreadful

It's not dreadful in a proper setup, the additional features over other filesystems also make it worthwhile. It can do tiered storage which BTRFS cannot presently. That is hot data is cached on faster storage layers, such that you can have HDDs for bulk cold storage and redundancy along with a SATA/NVMe SSD or something faster/low-latency like Optane 905P or RAM disk for your hot/frequent data.

If Michael did a proper benchmark showcasing that, you'd see other filesystems as absolutely dreadful in performance.

BTRFS will just eat all your data instead of a single file if it hits bitrot in its metadata. Not really an improvement.

If you have multiple copies of metadata like RAID1(single HDD is fine afaik, but SSD needs to be more than one disk since SSDs can optimize storage under the hood to share the same physical blocks/pages), that shouldn't happen?

When was the last case you can link to of this happening on a properly configured and maintained system? Only issues I've seen in past year or so is due to users enabling non-default features that usually aren't stable and cautioned against by the BTRFS wiki/docs. Plenty of reports of users that say BTRFS saved their data that would otherwise have been lost on other filesystems. More often than not, the data is recoverable on BTRFS.