• https://www.percona.com/blog/taking-a-look-at-btrfs-for-mysql/
  • zfs outperforms immensely, but potential misconfiguration on btrfs side (virt+cow still enabled?)
• https://www.ctrl.blog/entry/btrfs-vs-ext4-performance.html
  • see the follow up comment on this post

    • https://www.reddit.com/r/archlinux/comments/o2gc42/is_the_performance_hit_of_btrfs_serious_is_it/

      I’m the author of OP’s first link. I use BtrFS today. I often shift lots of de-duplicatable data around, and benefit greatly from file cloning. The data is actually the same data that caused the slow performance in the article. BtrFS and file cloning now performs this task quicker than a traditional file system. (Hm. It’s time for a follow-up article.)

      In a laptop with one drive: it doesn’t matter too much unless you do work that benefit from file cloning or snapshots. This will likely require you to adjust your tooling and workflow. I’ve had to rewrite the software I use every day to make it take advantage of the capabilities of a more modern file system. You won’t benefit much from the data recovery and redundancy features unless you’ve got two storage drives in your laptop and can setup redundant data copies.

      on similar hardware to mine?

      It’s not a question about your hardware as much as how you use it. The bad performance I documented was related to lots and lots of simultaneous random reads and writes. This might not be representative of how you use your computer.

• https://dl.acm.org/doi/fullHtml/10.1145/3386362
  • this is about distributed file systems (in this case Ceph) - they argue against basing DFS on ondisk-format filesystems (XFS ext4) - developed BlueStore as backend, which runs directly on raw storage hardware.
  • this is a good approach, but expensive (2 years in development) and risky
  • better approach is to take advantage of a powerful enough existing ondisk-FS format and pair it with supporting modules which abstract away the 'distributed' mechanics.
  • the strategy presented here is critical for enterprise-grade hardware where the ondisk filesystem becomes the bottleneck that you're looking to optimize
• https://lore.kernel.org/lkml/cover.1676908729.git.dsterba@suse.com/
  • linux 6.3 patch by David Sterba [2023-02-20 Mon]
  • btrfs continues to show improvements in the linux kernel, ironing out the kinks
  • makes it hard to compare benchmarks tho :/

• see this WIP k-ext for macos: macos-btrfs
  • maybe we can help out with the VFS/mount support

• on-disk-format
• 'btrfs consists entirely of several trees. the trees use copy-on-write.'
• trees are stored in nodes which belong to a level in the b-tree structure.
• internal nodes (inodes) contain refs to other inodes on the next level OR
  • to leaf nodes then the level reaches 0.
• leaf nodes contain various types depending on the tree.
• basic structures
  • 0:8 uint = objectid, each tree has its own set of object IDs
  • 8:1 uint = item type
  • 9:8 uint = offset, depends on type.
  • little-endian
  • fields are unsigned
  • superblock
    • primary superblock is located at 0x10000 (64KiB)
    • Mirror copies of the superblock are located at physical addresses 0x4000000 (64 MiB) and 0x4000000000 (256GiB), if valid. copies are updated simultaneously.
    • during mount only the first super block at 0x10000 is read, error causes mount to fail.
    • BTRFS onls recognizes disks with a valid 0x10000 superblock.
  • header
    • stored at the start of every inode
    • data following it depends on whether it is an internal or leaf node.
  • inode
    • node header followed by a number of key pointers
    • 0:11 key
    • 11:8 uint = block number
    • 19:8 uint = generation
  • lnode
    • leaf nodes contain header followed by key pointers
    • 0:11 key
    • 11:4 uint = data offset relative to end of header(65)
    • 15:4 uint = data size
• objects
  • ROOT_TREE
    • holds ROOT_ITEMs, ROOT_REFs, and ROOT_BACKREFs for every tree other than itself.
    • used to find the other trees and to determine the subvol structure.
    • holds items for the 'root tree directory'. laddr is store in the superblock
  • objectIDs
    • free ids: BTRFS_{FIRSTFREEOBJECTID}=256ULL:BTRFS_{LASTFREEOBJECTID}=-256ULL
    • otherwise used for internal use

• send stream format
• Send stream format represents a linear sequence of commands describing actions to be performed on the target filesystem (receive side), created on the source filesystem (send side).
• The stream is currently used in two ways: to generate a stream representing a standalone subvolume (full mode) or a difference between two snapshots of the same subvolume (incremental mode).
• The stream can be generated using a set of other subvolumes to look for extent references that could lead to a more efficient stream by transferring only the references and not full data.
• The stream format is abstracted from on-disk structures (though it may share some BTRFS specifics), the stream instructions could be generated by other means than the send ioctl.
• it's a checksum+TLV
• header: u32len,u16cmd,u32crc32c
• data: type,length,raw data
• the v2 protocol supports the encoded commands
• the commands are kinda clunky - need to MKFIL/MKDIR then RENAM to create

• magic#: 0x94
  • https://docs.kernel.org/userspace-api/ioctl/ioctl-number.html
  • Btrfs filesystem some lifted to vfs/generic
  • fs/btrfs/ioctl.h and linux/fs.h

– ?? (a)

Zoned Storage is an open source, standards-based initiative to enable data centers to scale efficiently for the zettabyte storage capacity era. There are two technologies behind Zoned Storage, Shingled Magnetic Recording (SMR) in ATA/SCSI HDDs and Zoned Namespaces (ZNS) in NVMe SSDs.

– zonedstorage.io – $465 8tb 2.5"? retail

• ioctl-numbers

• crates.io

• crates.io

• crates.io

• crates.io

• crates.io

• crates.io

• crates.io

• crates.io

• crates.io

• tracking-issue

• tracking-issue

• github

• github

• github

• github

• github

• github

• github

• github

• github

• github