Announcement

Collapse
No announcement yet.

Fedora Developers Evaluating Compression Options For Btrfs-By-Default Proposal

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • RomuloP
    replied
    Originally posted by skeevy420 View Post

    I wish I could remember where I read this, but with BTRFS+Zstd you want to exclusively use "compress-force" because "compress" will use the kernel's built-in heuristic to skip files where as "compress-force" will use Zstd's faster heuristic.
    In fact kernel heuristic was fast but poor. I'm not sure this is a problem since kernel 4.15. Kernel heuristic nowadays is quite sophisticated and probably as fast according to btrfs wiki, only bench-marking to be sure it is worth it.
    Last edited by RomuloP; 09 July 2020, 12:20 PM.

    Leave a comment:


  • skeevy420
    replied
    Originally posted by RomuloP View Post

    BTRFS does the same per file, zstd can skip in-compressible data automatically. And the heuristic is pretty fast, more than 1GB/s per thread. To the point you have `compress` and `compress-force`
    I wish I could remember where I read this, but with BTRFS+Zstd you want to exclusively use "compress-force" because "compress" will use the kernel's built-in heuristic to skip files where as "compress-force" will use Zstd's faster heuristic.

    Leave a comment:


  • antonyshen
    replied
    I use BTRFS w/ zstd:3 compression on Ubuntu, including a rpi4 installation. I love BTRFS, it's much much more usable that before, offers lots of useful features.

    Leave a comment:


  • RomuloP
    replied
    Originally posted by k1e0x View Post
    FreeBSD's been doing this for 10 years or better now.. They just pretty much turn on lz4 for almost everything non base due to lz4 skipping files it can't compress quickly.
    BTRFS does the same per file, zstd can skip in-compressible data automatically. And the heuristic is pretty fast, more than 1GB/s per thread. To the point you have `compress` and `compress-force`





    Leave a comment:


  • RomuloP
    replied
    The only thing I miss implementation is encryption and the only thing I miss a fix is quotas... Quotas have a huge performance impact but is a relevant feature.

    Leave a comment:


  • RahulSundaram
    replied
    Originally posted by ezst036 View Post
    Is there any reason why xfs isn't the default?

    I would be perfectly happy with btrfs, it just seems to me that xfs provides the path of least resistance away from ext4.
    XFS is the default for Fedora server. For workstation, shrinking was considered important and the feature set for Btrfs is quite different

    Leave a comment:


  • ezst036
    replied
    Is there any reason why xfs isn't the default?

    I would be perfectly happy with btrfs, it just seems to me that xfs provides the path of least resistance away from ext4.

    Leave a comment:


  • ThoreauHD
    replied
    They really need to not F that up. Torvalds uses that OS, and CoC or not, he'll rip them if they don't test reality.
    Last edited by tildearrow; 08 July 2020, 04:41 PM.

    Leave a comment:


  • k1e0x
    replied
    FreeBSD's been doing this for 10 years or better now.. They just pretty much turn on lz4 for almost everything non base due to lz4 skipping files it can't compress quickly.

    Dataset compression settings:
    zpool compression off
    zpool/ROOT compression off
    zpool/ROOT/initial compression lz4
    zpool/root compression lz4
    zpool/tmp compression lz4
    zpool/usr compression off
    zpool/usr/home compression lz4
    zpool/usr/jails compression lz4
    zpool/usr/obj compression lz4
    zpool/usr/ports compression lz4
    zpool/usr/src compression lz4
    zpool/var compression off
    zpool/var/audit compression lz4
    zpool/var/log compression lz4
    zpool/var/mail compression lz4
    zpool/var/tmp compression lz4

    That is a pretty simple laout, binaries are left uncompressed (inherit from /usr) If you're looking for a model.. that is a good start. Ubuntu I know has taken a more varied layout but it's similar.
    Last edited by k1e0x; 08 July 2020, 03:38 PM.

    Leave a comment:


  • cjcox
    replied
    Thin provisioning, over provisioning and/or compression == future "surprise" failure.

    Leave a comment:

Working...
X