Btrfs Seeing Some Nice Performance Improvements For Linux 5.9

[S.Pam]

Hot data tracking/balancing is already considered

[darkbasic]

What distro? Ubuntu is known for btrfs horror stories. I've been following btrfs mailing list and a lot of the people asking for help are from some random Ubuntu LTS using older kernels.

In other distros like OpenSUSE where btrfs is default filesystem it doesn't blow up like that. Now also on Fedora will become first-class citizen.

The same happened to me ~5 years ago with Arch Linux, but I still use btrfs because there are no better alternatives.

[vladpetric]

you mean for benchmarking speed of filesystem resize?

Some of the benchmarks (e.g., sqlite, postgres) use fsync ... The degree to which they are sped up is TBD.

[cynical]

This makes me wonder whether I should use ZFS or Btrfs for my next Ubuntu install. I’m leaning towards ZFS because the official support is likely to be stronger.

[starshipeleven]

The same happened to me ~5 years ago with Arch Linux, but I still use btrfs because there are no better alternatives.

Well, if it didn't happen again it's OK I guess?

[starshipeleven]

This makes me wonder whether I should use ZFS or Btrfs for my next Ubuntu install. I’m leaning towards ZFS because the official support is likely to be stronger.

+1 for ZFS, Ubuntu or not.

[dev_null]

What distro? Ubuntu is known for btrfs horror stories. I've been following btrfs mailing list and a lot of the people asking for help are from some random Ubuntu LTS using older kernels.

In other distros like OpenSUSE where btrfs is default filesystem it doesn't blow up like that. Now also on Fedora will become first-class citizen.

gentoo.

[theriddick]

I'm going to setup btrfs on my 2TB NVMe and test it under windows (free driver) and linux. Will be interesting, assuming my drive even turns up from China..................

[pal666]

Right now there is not much of a difference, but in the long run this could allow for parity based raid where the stripes are on fast groups of disks and the parity on slower devices. And a disk failure could use a different group as spare space or redundant space. It works also make it easier to isolate metadata on its own controller for example. Plus migrationomigration data would not be across filesystems. It all boils down to flexibility.

nothing you listed requires it to be subvolume-specific. i.e. if you could put parity for whole fs on separate drives, it would work just as well(because all good things about subvolumes stem from single fact that they are not tied to partitions. as soon as you tie them, you don't need them)

[pal666]

Some of the benchmarks (e.g., sqlite, postgres) use fsync ... The degree to which they are sped up is TBD.

some people are smart enough to not use postgres on cow fs. less smart ones look for bogus benchmarks